Livestock Ration Formulation For Dairy Cattle and Buffalo CRC Press

https://t.
me/livestockscience
Livestock Ration
Formulation for Dairy
Cattle and Buffalo
e
nc
ie
sc
k
oc
st
ve
li
e/
t.m
s://
tp
ht
https://t.me/livestockscience
e
nc
ie
sc
ck
to
es
liv
e/
.m
//t
s:
tp
ht
Livestock Ration
Formulation for Dairy
Cattle and Buffalo
Ravinder Singh Kuntal

Assistant Professor, Department of Mathematics, Faculty of Engineering
and Technology, Jain (Deemed-to-Be) University, Bengaluru, Karnataka, India
Radha Gupta
Department of Mathematics, Dayananda Sagar College of Engineering,
Bengaluru, Karnataka, India
D. Rajendran
ICAR-National Institute of Animal Nutrition and Physiology, Bengaluru,
Karnataka, India
Vishal Patil
Department of Mathematics, Faculty of Engineering and Technology, Jain
(Deemed-to-be) University, Bengaluru, Karnataka, India
MATLAB® is a trademark of The MathWorks, Inc. and is used with permission. The
MathWorks does not warrant the accuracy of the text or exercises in this book. This
book’s use or discussion of MATLAB® software or related products does not constitute
endorsement or sponsorship by The MathWorks of a particular pedagogical approach or
particular use of the MATLAB® software.
First edition published 2022
by CRC Press
6000 Broken Sound Parkway NW, Suite 300, Boca Raton, FL 33487–2742
and by CRC Press
2 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN
© 2022 Ravinder Singh Kuntal, Radha Gupta, Rajendran D and Vishal Patil
CRC Press is an imprint of Taylor & Francis Group, LLC
Reasonable efforts have been made to publish reliable data and information, but the
author and publisher cannot assume responsibility for the validity of all materials or the
consequences of their use. The authors and publishers have attempted to trace the copyright
holders of all material reproduced in this publication and apologize to copyright holders if
permission to publish in this form has not been obtained. If any copyright material has not
been acknowledged please write and let us know so we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted,
reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means,
now known or hereafter invented, including photocopying, microflming, and recording, or in
any information storage or retrieval system, without written permission from the publishers.
For permission to photocopy or use material electronically from this work, access www.
copyright.com or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood
Drive, Danvers, MA 01923, 978–750–8400. For works that are not available on CCC please
contact mpkbookspermissions@tandf.co.uk
Trademark notice: Product or corporate names may be trademarks or registered trademarks
and are used only for identifcation and explanation without intent to infringe.
Library of Congress Cataloging‑in‑Publication Data
Names: Singh Kuntal, Ravinder, author. | Gupta, Radha, 1970– author. |
Rajendran, D., author. | Patil, Vishal, author.
Title: Livestock ration formulation for dairy cattle and buffalo /
Ravinder Singh Kuntal, Radha Gupta, Rajendran D, Vishal Patil.
Description: First edition | Boca Raton, FL : CRC Press, 2022. |
Includes bibliographical references and index.
Identifers: LCCN 2021048758 | ISBN 9781032137476 (hardback) |
ISBN 9781032139678 (paperback) | ISBN 9781003231714 (ebook)
Subjects: LCSH: Dairy cattle—Feeds and feeding. | Buffaloes—Feeds and
feeding. | Dairy farming. | Animal nutrition. | Handbooks and manuals.
Classifcation: LCC SF95 .S564 2022 | DDC 636.08/52—dc23/eng/20211122
LC record available at https://lccn.loc.gov/2021048758
ISBN: 978-1-032-13747-6 (hbk)
ISBN: 978-1-032-13967-8 (pbk)
ISBN: 978-1-003-23171-4 (ebk)
DOI: 10.1201/9781003231714
Typeset in Times
by Apex CoVantage, LLC
This book is dedicated to the memory of my beloved father
Late Warrant Offcer R.L Verma (1956–1999)
&
To my mother
Smt. Premvati Devi Verma
You are profoundly appreciated for your unfinching
commitment to the pursuit of my career
And also, to my son, Vivaan
This book was written when you were separated from me,
and I just want to tell you that I love you so much.
e
nc
ie
sc
ck
to
es
liv
e/
.m
//t
s:
tp
ht
Contents
List of Tables........................................................................................................... xi
List of Figures.......................................................................................................xiii
Preface ................................................................................................................... xv
Acknowledgments................................................................................................xvii
Authors..................................................................................................................xix
List of Abbreviations ..........................................................................................xxiii
1. Introduction to Animal Nutrition................................................................. 1

1.1 Operational Research............................................................................ 1
1.2 Animal Nutrition................................................................................... 2
1.3 Essential Nutrients ................................................................................ 3
1.3.1 Energy ..................................................................................... 4
1.3.2 Protein ..................................................................................... 4
1.3.3 Minerals .................................................................................. 4
1.4 Classifcation of Feedstuffs................................................................... 5
1.5 Application Area................................................................................... 6
2. Review of Literature ...................................................................................... 9

2.1 Introduction........................................................................................... 9
2.2 Linear Programming in Ration Formulation...................................... 10
2.2.1 Linear Programming Based Software .................................. 13
2.2.2 Limitations of Linear Programming in Ration
Formulation........................................................................... 13
2.2.3 Overcoming the Limitations of Linear Programming ........... 14
2.3 Nonlinear Programming..................................................................... 14
2.4 Goal Programming ............................................................................. 18
2.5 Genetic Algorithm .............................................................................. 19
3. Tools and Techniques................................................................................... 21

3.1 Computation of LP Model Using MS-Excel Based Solver................. 21
3.1.1 Simplex LP Method in MS-Excel Solver.............................. 22
3.1.2 GRG Nonlinear in MS-Excel Solver..................................... 22
3.1.3 Evolutionary Algorithm in MS-Excel Solver ....................... 22
3.1.4 MATLAB (Matrix-Laboratory)............................................ 23
3.1.4.1 Inputs of Genetic Algorithm................................. 24
3.2 Data Handling and Nutrient Bound Calculator .................................. 26
4. Binary Coded Genetic Algorithm to Solve Ration

Formulation Problem................................................................................... 27
4.1 Introduction......................................................................................... 27
vii
viii Contents
4.2 Linear and Nonlinear Programming Model for Animal Feed

Formulation (Test Problem-1) ............................................................. 28
4.2.1 Linear Programming Formulation........................................ 28
4.2.2 Nonlinear Programming Formulation .................................. 29
4.2.3 Extended Work Using Genetic Algorithms .......................... 29
4.2.4 Solution of Nonlinear Programming Model by
Genetic Algorithm ................................................................ 30
4.2.5 Graphical Results.................................................................. 32
4.3 Goal Programming Model (Test Problem-2)...................................... 33
4.3.1 Formulation of the Nonlinear Goal Programming
Problem ................................................................................. 33
4.3.2 Graphical Results.................................................................. 39
4.3.3 Conclusion............................................................................. 42
5. Least Cost Feed Formulation for Dairy Cattle During Pregnancy

by Using Real Coded Genetic Algorithm: An Application ...................... 45
5.1 Introduction......................................................................................... 45
5.1.1 Data Collection and Methodology ........................................ 46
5.1.1.1 LP Model for Cattle-1........................................... 50
5.2 Implementation ................................................................................... 52
5.2.1 Results and Discussions ........................................................ 53
5.2.2 Conclusion............................................................................. 59
5.3 A Goal Programming Approach to Ration Formulation
Problem for Indian Dairy Cows.......................................................... 60
5.3.1 Goal Programming Approach .............................................. 62
5.3.1.1 GP Model-1........................................................... 63
5.3.1.2 GP Model-2........................................................... 63
5.3.1.3 GP Model-3........................................................... 64
5.3.2 Real Coded Genetic Algorithm with Hybrid Function......... 65
5.3.3 Results................................................................................... 66
5.3.4 Discussion ............................................................................. 67
5.3.5 Conclusion............................................................................. 68
6. Study of Real Coded Hybrid Genetic Algorithm to Find Least

Cost Ration for Non-Pregnant Dairy Buffaloes ........................................ 69
6.1 Introduction......................................................................................... 69
6.2 Data Collection and Data Handling.................................................... 72
6.3 Problem Description ........................................................................... 73
6.3.1 Linear Programming Model ................................................. 74
6.3.2 Goal Programming Model-1................................................. 75
6.3.3 Goal Programming Model-2................................................. 76
6.3.4 Goal Programming Model-3 (Auxiliary Approach)............. 76
6.4 Development of Real Coded Hybrid Genetic Algorithm ................... 77
Contents ix
6.4.1 Objective Function, Decision Variables, Representation

of Plots .................................................................................. 77
6.4.2 Crossover, Selection, Mutation and Elitism Operators......... 78
6.4.3 Global vs. Local Minima ...................................................... 78
6.5 Results on Dry Matter and As-Fresh Basis ........................................ 79
6.6 Discussion ........................................................................................... 80
6.7 Conclusion........................................................................................... 86
7. Conclusion..................................................................................................... 87
References............................................................................................................. 95
Index .................................................................................................................... 103
e
nc
ie
sc
ck
to
es
liv
e/
.m
//t
s:
tp
ht
Tables
Table 1.1 Nutrient Content of Common Feeds and Fodder for Dairy
Cattle and Buffalo on DM Basis ........................................................ 6
Table 4.1 Composition of Concentrate Mixtures in Respect of DCP
and TDN ........................................................................................... 28
Table 4.2 Solution for NLP Model by GA for Original Bounds
(Test Problem-1) ............................................................................... 30
Table 4.3 Solution for NLP Model by GA for Reduced Bounds
(Test Problem-2) ............................................................................... 31
Table 4.4 Comparison of Solution Obtained by GA and RST for
NLP Model ....................................................................................... 31
Table 4.5 Nonlinear GP Model with Priority Ranked Goals for Animal
at Level-1 to 3 with Objective Function (z) to Different Goals
Which Need to Be Minimized ......................................................... 35
Table 4.6 Solution Set of LPP for Level-1, 2 and 3 by GA .............................. 36
Table 4.7 Solution Set of the Nonlinear Programming Problem (NLPP)
for Level-1, 2 and 3 by GA................................................................ 37
Table 4.8 Comparison of Decision Variables by RST and GA........................ 38
Table 5.1 Nutrient Requirements of Cattle-1, 2 and 3 ..................................... 47
Table 5.2 Selected Feedstuffs........................................................................... 47
Table 5.3 Nutrient Content of Feedstuffs on DM Basis................................... 48
Table 5.4 Least Cost of Ration by Various Techniques for Cattle-1
on DM Basis ..................................................................................... 54
on As Fresh Basis (LP Model) ......................................................... 54
on DM Basis (LP Model) ................................................................. 55
on As Fresh Basis (LP Model) ......................................................... 56
on DM Basis (LP Model) ................................................................. 56
xi
xii Tables

on As Fresh Basis ............................................................................. 57
Table 5.10 Net Proft Per Cattle in INR to Farmers for 10 Litres of Milk
on As Fresh Basis Each Day ............................................................ 58
Table 5.11 Least Cost, Deviation, and Constraints Value Solved by
Hybrid RGA for Goal Programming Models................................... 66
Table 6.1 Nutrient Requirements for Non-Pregnant Buffalo Weighing
450 kg and Yielding 10 Litres of Milk with 6% Fat ........................ 72
Table 6.2 Nutrient Content of Feedstuffs on DM Basis................................... 73
Table 6.3 Priority Values for GP Model-1........................................................ 76
Table 6.4 Priority Values for GP Model-2 ....................................................... 76
Table 6.5 Least Cost Ration Formulated by Various Techniques for
Non-Pregnant Dairy Buffalo on DM Basis (LP Model) .................. 79
Table 6.6 Least Cost Ration Formulated by Various Techniques
for Non-Pregnant Dairy Buffalo on “As Fresh Basis”
(LP Model) ....................................................................................... 81
Table 6.7 Least Cost and Deviation Value Solved by Hybrid RGA for
GP Model-1 ...................................................................................... 82
GP Model-2 ...................................................................................... 83
GP Model-3 ...................................................................................... 84
Figures
Figure 4.1 Maximum Milk Yield Using GA (Original Bounds). ...................... 32

Figure 4.2 Maximum Milk Yield Using GA (Reduced Bounds). ..................... 32
Figure 4.3 Maximum Milk Yield Using RST and GA (Original Bounds). ..........32
Figure 4.4 Maximum Milk Yield Using RST and GA (Reduced Bounds). .........33
Figure 4.5 Graphs of Deviations for Linear Model of Level-1, 2 and 3. ........... 39
Figure 4.6 Deviations for Nonlinear GP Model of Level-1, 2 and 3. ................ 40
Figure 4.7 Comparison of GA and RST. ........................................................... 41
Figure 5.1 Schematic Diagram Representing Methodology Followed
for Least Cost Feed Formulation of Dairy Cattle. ........................... 49
Figure 6.1 Methodology to Calculate Least Cost Balanced Ration for
Non-Pregnant Dairy Buffalo............................................................ 74
xiii
e
nc
ie
sc
ck
to
es
liv
e/
.m
//t
s:
tp
ht
Preface
Optimization is one of the important areas of mathematics and computer science in

which linear programming (LP) and goal programming (GP) are two workhorses
which provide solutions to the user. The concept of LP and GP can be applied in
animal farming to formulate the least cost balanced ration. Finding the least cost
ration which satisfes the animals’ need is important because ration cost is the single
greatest variable which determines the proftability of dairy farming. Dairy farm-
ers in the Mandya district of Karnataka have very limited feed resources to fulfll
the nutrients requirement in terms of crude protein (CP), total digestible nutrients
(TDN), calcium (Ca) and phosphorus (P); therefore, one needs a software program
to plan a balanced low budget diet. Linear programming and goal programming
approaches, alone or in combination, have proved benefcial for fnding solutions
to ration formulation and diet-related problems of many animals and poultry. This
study is a step forward in that direction to provide least cost diet formulation based
on nutrient requirements of cattle and buffalo, which has been calculated according
to the specifcations of the Indian Council of Agricultural Research (ICAR, 2013)
and National Research Council (NRC, 2001). Linear programming models for three
types of “dairy cattle with body weight 500 kg each and requirement of 10 litre milk
yield with 4% of fat content with seventh, eight, and ninth months of pregnancy”
and “non-pregnant dairy buffalo weighing 450 kg and yielding 10 L milk with 6%
of fat content” are considered on a dry matter basis. Further, a primitive goal pro-
gramming model for three categories of dairy cattle and non-pregnant buffalo has
been formulated by categorizing the goals into a set of priorities. Goal programming
models have been developed because there are two high priority objectives: least
cost and the dry matter intake to be achieved (if possible). These linear program-
ming models for dairy cattle are comprehended by generalized reduced gradient
(GRG) nonlinear, evolutionary, simplex linear programming (LP) and real coded
genetic algorithm (RGA) with and without seeding the random number generation
methods. The outcome reveals there was no signifcance difference (P > 0.05) found
with various techniques adopted for feed preparation, and RGA can be adopted for
feed conceptualization. The linear and goal programming model for non-pregnant
dairy buffalo was solved using hybrid real coded genetic algorithm, and the results
are compared with (RGA), considering different versions like RGA without cross-
over, RGA without mutation and RGA with crossover and mutation. Here it is
observed, however, that RGA without crossover and RGA without mutation operator
provide a near to optimal answer, but solutions seem to get stuck in local minima.
Hence, RGA with hybrid function provides the optimal solution, and this method
can be utilized for cheap feed formulation for dairy cows and buffaloes. This study
reveals that the rations of dairy animals can be optimized using real coded genetic
algorithm by solving linear and goal programming models of feed formulation.
xv
Acknowledgments
Shrii Saraswatii Namahstubhyam Varade Kaama Ruupini Twaam Aham

Praarthane Devii Vidyaadaanam Cha Dehi Me
“I bow to Goddess Saraswati, who fulfls the wishes of the devotees. I pray to her
to enlighten me with knowledge. Saraswati is the provider of boons and the one
who grants all our desires. She is capable of removing ignorance and bestowing
intelligence. I pray to the Goddess to help me in making this work fruitful and
make me successful in all my efforts.” I am grateful to Lord Ganesha for giving me
patience and strength to overcome diffculties, which helped me crossing my way
in the accomplishment of this endeavour.
It is hard to say whether it has been grappling with the topic itself which has been
the real learning experience or grappling with how to write papers, give talks, work
in a group, stay up until the birds start singing and stay focused. In any case, I am
indebted to many people for making the time working on my book an unforgettable
experience. It would not have been possible to write this book without the help and
support of the kind people around me. I acknowledge many known and unknown
hands that have helped me in this process. No words could adequately express my
gratitude to them, and I shall ever remain indebted to them all. I express my heart
with a profound sense of gratitude to my supervisor Dr. Radha Gupta, Professor
and Head, Department of Mathematics, Dayananda Sagar College of Engineering,
Bangalore. She has oriented and supported me with promptness and care and has
always been patient and encouraging in times of new ideas and diffculties. Her
excellent quality in teaching the subject will remain in my mind forever. Her hard
work, constant patience and determined mind set an example. This book would
not have been possible without her help, constant support and patience. Her invalu-
able help of constructive comments and suggestions throughout my research work
have contributed to the progress of this research. I am very much thankful to
Dr. D. Rajendran, Principal Scientist, Animal Nutrition Division, National Institute
of Animal Nutrition and Physiology (NIANP), Adugodi, Bangalore, for provid-
ing primary data and continuous guidance in this research journey. His enormous
research experience as well as subject knowledge is source of inspiration for me
which helped me to carry this work to this level. He also helped me in improv-
ing my writing skills, fnding appropriate journals and in making an Excel-based
nutrient bound calculator for calculating nutrient requirements of dairy cattle and
buffaloes and result validation. Above all, he made me feel as a friend, which I
appreciate from my heart. I am deeply grateful to Dr. Sandeep Shastri, Professor
and the Vice Chancellor of Vice Chancellor, Jagran Lake City University, Bhopal,
who has made me understand the word research. His innovative and systematic
way of teaching research methodology has helped me to carry out the work in an
effective way. He has provided the right direction to the entire path of research and
xvii
xviii Acknowledgments
enlightened and entertained us with his knowledge, experience and skill in this
feld. I can’t imagine my current position without the support from my mother and
sister. A formal statement of acknowledgment will hardly meet the end of justice
in the matter of the expression of a deep sense of gratitude to my mother, Smt.
Premvati Devi. I express my deep sense of love and gratitude for her great patience
at all times, constant encouragement and inspiration. I also express my deep sense
of gratitude to Dr. Anita Chaturvedi, Professor, Department of Mathematics, FET
Jain (Deemed-to-be) University for her kind support and encouragement at all times.
I am also very much thankful to Dr. Kokila Ramesh, Department of Mathematics,
FET Jain (Deemed-to-be) University for extending her help in times of need. I
also express my deep sense of gratitude to Ms. Priya Chakraborty, Department of
English, FET Jain (Deemed-to-be) University for extending her help in correcting
grammatical errors in this text book.
Authors
Dr. Radha Gupta earned her M.Sc. degree in mathematics from

Agra University, Uttar Pradesh, India, in 1992 and M.Phil. degree
in mathematics from the University of Roorkee, Uttarakhand,
India, in 1993. She earned her Ph.D. degree in mathematics from
Vikram University, Ujjain, Madhya Pradesh, India, in 2011.
From 1993 to 1996, she worked as a research fellow with the
Central Building Research Institute, Roorkee, Uttarakhand.
From 1999 to 2004, she was a lecturer in SVM PU College
Bangalore; from 2004 to 2007 she was a lecturer in the PES
Institute of Technology, Bangalore. She moved to the School of Engineering and
Technology, Jain University, Bangalore, in 2007 and served the Institution as
Associate Professor and Head, Department of Basic Science, till 2015. Since 2015,
she has been a professor and department head with the Mathematics Department,
Dayananda Sagar College of Engineering, Bangalore. Her research interests
include operations research, differential equations, mathematical modelling and
statistical analysis.
Dr. Radha Gupta is the author of fve books, some conference papers and more
than 50 research publications in reputed national and international journals. She is
also a reviewer for some prestigious journals. She is a recognized research super-
visor from Visvesvaraya Technological University (VTU), Belagavi and a member
of many professional bodies. Dr. Radha Gupta was a recipient of the President
Gold Medal (from then President of India, Shri Shankar Dayal Sharma) for the
best student of the year, 1992, Agra University, Uttar Pradesh, and also a recipient
of the University Gold Medal for standing frst class frst in M.Phil. (mathematics)
in the year 1993, University of Roorkee, Uttarakhand.
Dr. Ravinder Singh Kuntal is an expert in optimizing and creat-

ing user-friendly applications in Microsoft Excel and the Android
platform for least cost ration formulation for dairy animals. His
research interests include operations research, data analytics,
mathematical modelling and statistical analysis. He has worked
on these models after eight years of experience in least cost opti-
mization research. He is currently an Assistant Professor of
Mathematics in the Faculty of Engineering and Technology, Jain (Deemed-to-be
University) Bangalore. He has a total of 11 years of teaching and eight years of
research experience. He has published 14 research papers in peer-reviewed jour-
nals. He has received many awards for his research work and has bagged the title of
Best Paper Award for “Least Cost Formulation for Ruminants” in the International
Conference on Applications of Artifcial Intelligence and Computational
Mathematics (AAICM – 2020). One of his publications also received an
xix
xx Authors
acknowledgment of Best Paper Award on “Study of Real Coded Hybrid Genetic

Algorithm (RGA) to Find Least Cost Ration for Non-Pregnant Dairy Buffaloes” at
the 7th International conference on Soft Computing for Problem Solving
(SocPros–2017) December 23–24, 2017, at IIT Bhubaneswar, the Best Paper Award
titled “heuristic approach to nonlinear optimization technique” at Research Retreat:
Exploring Pathways, Unlocking Ideas” at Jain University. He was named the Best
Mathematics Professor of the year in recognition of continuous excellence in teach-
ing, awarded at International Education Awards conference 2020. He has a research
collaboration with the NIANP (National Institute of Animal Nutrition and
Physiology) Bangalore, India. He also completed consultancy projects for ICAR-
NIANP institutions and “Design and Development Least Cost Feed Formulation
(LCFF) Application for Goat, Sheep & Camel Using Excel VBA”, for King Saud
University, Riyadh, Saudi Arabia.
Dr. Vishal Patil is an expert in developing and designing least

cost linear programming (LP) and stochastic models and creating
user-friendly application in Microsoft Excel and the Android plat-
form. He has built these models after ten years of experience in
teaching and research. He has completed consultancy projects for
ICAR-NIANP institutions and “Design and Development Least
Cost Feed Formulation (LCFF) Application for Goat, Sheep &
Camel Using Excel VBA”, for King Saud University, Riyadh, Saudi Arabia. He has
published ten peer-reviewed international papers. He is an Assistant Professor of
Mathematics in the Faculty of Engineering & Technology, Jain (Deemed-to-be
University), Bangalore. His major contribution includes research collaboration with
NIANP (National Institute of Animal Nutrition and Physiology) Bangalore, India.
Professor D. Rajendran, Principal Scientist, ICAR-National

Institute of Animal Nutrition and Physiology, completed his
Bachelor of Veterinary Science at Tamilnadu Veterinary and
Animal Sciences University, Chennai, in 1997 and his Master
in Animal Nutrition at the Indian Veterinary Research Institute,
Bareilly. He was a recipient of the Junior Research Fellowship
Award during his master’s degree. In 2000 he joined West Bengal
University of Animal and Fishery Science as a lecturer. He pursued his doctoral
degree programme at Madras Veterinary College, Chennai. Professor Rajendran
received three Gold Medals during his PhD program, namely the TANUVAS Gold
Medal, Dr. Richard Masilamani Memorial Award and TANUVAS Alumni Award.
He joined as Assistant Professor in Animal Nutrition at the Veterinary College and
Research Institute, Namakkal, Taminadu and established the Feed Mill under a
revolving fund project. He conducted research on nano minerals and guided two
undergraduate students to receive the ALLTECH Young Scientist Award 2010 and
2011. He joined as senior scientist in 2010 at the ICAR-National Institute of Animal
Nutrition and Physiology, Bangalore. His pioneering work on nano technology
in animal nutrition for the improvement of bioavailability of mineral is an eye-
opener for the application of nano mineral science in animal nutrition. The work
Authors xxi
was recognized and received best thesis award by the Animal Nutrition Journal by
Elsevier Pvt. Ltd.
He has 185 publications, including 51 research articles, 12 book chapters, 37
technical articles and 20 teaching manuals. He received seven awards in con-
ferences and seminars of national and international repute. He was invited as
guest lecturer by CG Group at Boun Me Thout, DakLak, Vietnam. He success-
fully guided 12 master degree and fve doctoral degree students in the capacity
of chairman/member of the advisory committee. He developed four software and
two Android apps and published them for public use. He has published fve ready
reckoners and one diagnostic kit. More than 25000 copies of the ready reckoner
were sold and created a revolution in the feld of animal nutrition. He has success-
fully completed 11 externally/institute funded research projects, and currently has
fve projects in progress. The International Society of Biotechnology conferred the
Fellow of International Society of Biotechnology (FISBT) on him in recognition of
his contribution and achievements.
Abbreviations
b Bran Max
BGA Binary Coded Genetic Algorithm
Ca Calcium
c/h Legume/Non-Legume Green
Ck Cake Max
CP Crude Protein
d/g Dry/Green Roughage
DM Dry Matter
DMI Dry Matter Intake
EA Evolutionary Algorithm
GA Genetic Algorithm
gm/kg Gram per Kilogram
GP/GPP Goal Programming/Goal Programming Problem
GRG Generalized Reduced Gradient
ICAR Indian Council of Agricultural Research
INR Indian Rupees
KT Kuhn-Tucker
lc Least Cost
LP/LPP Linear Programming/Linear Programming Problem
NIANP National Institute of Animal Nutrition and Physiology
NLP/NLPP Nonlinear Programming/ Nonlinear Programming Problem
NRC National Research Council
P Phosphorus
r/c Roughage/Concentrate
RGA Real Coded Genetic Algorithm
RHGA Real Coded Hybrid Genetic Algorithm
RNG Random Number Generation
RST Controlled Random Search Technique
SQP Sequential Quadratic Programming
t Total (in Kg)
TDN Total Digestible Nutrients
TMR Total Mixed Ration
xxiii
e
nc
ie
sc
ck
to
es
liv
e/
.m
//t
s:
tp
ht
1
Introduction to Animal Nutrition
CONTENTS
1.1 Operational Research ..................................................................................... 1
1.2 Animal Nutrition ............................................................................................ 2
1.3 Essential Nutrients ......................................................................................... 3
1.3.1 Energy................................................................................................ 4
1.3.2 Protein ............................................................................................... 4
1.3.3 Minerals............................................................................................. 4
1.4 Classifcation of Feedstuffs ............................................................................ 5
1.5 Application Area ............................................................................................ 6
1.1 Operational Research

Operational research is a subject feld that comprehends a wide range of applica-
tions in logical methods to make effective decision-making. Further, operational or
operations research emerged while providing the solution to plan effcient military
operations during World War II. Operational research is used not only in military
operations but also in industry, government, agriculture and animal production.
Today’s operations research has overlap with many disciplines such as industrial
engineering, dairy etc., with the motto of maximizing the proft, performance or
yield as well as minimizing the loss, risk or cost of some real-world problems. We
cannot give any standard defnition to operations research, as its boundaries are
not fxed, but as per the Institute for Operations Research and the Management
Sciences (INFORMS):
Operations Research and the Management Sciences are the professional

disciplines that deal with the application of information technology for
informed decision-making.
Yet this explicit defnition is not suffcient, as it is different from a related feld.
However, operations research is a subfeld of applied mathematics. We are using
techniques from mathematics such as mathematical modeling, statistical analy-
sis and mathematical optimization to arrive at optimal or near-optimal solutions
to complex problems. Therefore, instead of following the INFORMS defnition,
we use operations research with the help of mathematical optimization to specify
DOI: 10.1201/9781003231714-1 1
2 Livestock Ration Formulation
techniques and solutions for specifc real-time problems. This method includes lin-
ear programming, nonlinear programming, integer programming, optimization,
the Markov process, queuing theory, goal programming and heuristic algorithms.
These techniques would be accepted because they give optimal solutions for spe-
cifc problems except for the fact that in linear programming, if the problem is not
defned in linear equations (equalities or inequalities) or continuous domain, then
we cannot apply linear programming algorithms. There are many techniques in
operations research which have proven to be very useful in practice. For many real-
world optimization problems, operations research provides the method of choice,
in particular for those problems that are complex in nature.
1.2 Animal Nutrition

India has the largest livestock population in the world. Livestock is one of the most
important economic activities as it plays an important role in growth of Indian
agriculture, the growth of Indian economy (it contributes about 4% of national
gross domestic product – GDP – and 25% on agricultural GDP) and in livelihood,
especially in the rural areas of the country, providing income for most of the fam-
ily. In dairy farming, feeding cost accounts for about 70% of total operation cost.
As per the Government of India Ministry of Agriculture Department of Animal
Husbandry, Dairying, and Fisheries, Krishi Bhavan, in the 19th livestock census-
2012 all India report, the total milch animals population (cows and buffaloes) has
increased to 118.59 million (2012), an increase of 6.75%. The female buffalo popu-
lation has increased by 7.99% over the previous census, and the total number of
female buffaloes was 92.5 million in 2012. The buffalo population increased from
105.3 million to 108.7 million, showing a growth of 3.19%. The population of cows
increased by 6.52%, and the total number of cows estimated in 2012 was 122.9
million. As per the Basic Husbandry & Fisheries statistics 2017, the per capita
availability of average milk in Karnataka was 291 grams per day during 2016–17,
which is less than the 12 top milk-producing states in India, such as Uttar Pradesh.
Karnataka has only a 4% share in milk production in year 2016–17. From 2012 to
2016, the cattle population increased from 1.14 million to 1.37 million, with esti-
mated milk production of 5718.22 to 6562.15 (in thousands) in which the average
yield per in-milk animal of nondescript/indigenous cows during 2012–13 to 2016–
17 in Karnataka was 2.32–2.43 kg/day. Area under fodder crops is increased from
35000 hectares in 2006–07 to 36000 hectares, and permanent pastures and other
grazing lands decreased from 930000 hectares to 906000 hectares from 2006–07
to 2013–14 (Basic Animal Husbandry & Fisheries Statistics, 2017). According to
Basic Animal Husbandry & Fisheries Statistics 2017 there was a decrease of the
cattle population (190 million vs. 199 million) but an increase in buffalo (108 mil-
lion vs. 105 million) population from the previous census (2012 vs. 2007). These
animals contribute about 165.4 million tons of milk production annually (2016–17),
in which the major contribution was from buffalo, greater than 49%. Per capita
availability of milk in 2010–11 increased from 281 g/d to 355 g/d in 2016–17. In
Introduction to Animal Nutrition 3
addition, during 2016–17, the share of milk production from cross-breed cows,
indigenous cows, nondescript cows and buffaloes was 25.4%, 11.3%, 9.5% and
49.2%, respectively, of total milk production. The increase in population of exotic/
cross-breed cattle and indigenous cattle was 34.8 % and 0.17%, respectively,
whereas the population of milk animals (cows and buffaloes) has increased from
77.04 million to 80.52 million. Increasing the productivity of animals is one of the
major issues in our country, which can be overcome by developing proper feeding
systems to provide balanced nutrients to fulfll their requirements. By looking at
increasing demands of animal products such as milk and dairy products, we can
see a growth in a new entrepreneurship area with a commercial outlook for which
scientifc management and sustainability are important components. Nutrients
are very important to dairy animals to maintain their body conditions, for milk
production with an adequate amount of fat percentage and for maintaining preg-
nancy at third trimester. These nutrients are provided through feed, and the nutrient
requirements are measured in terms of energy, protein, minerals and vitamins. In
many developed countries, the standard feeding practice has come through various
continuous experiments, evaluation and refnement for decades. But in India, the
frst published document on scientifc feeding for Indian cattle appeared in ICAR
bulletin No 25, titled “Nutritive Values of Indian Cattle Feeds and the Feeding of
Animals” by Sen (1957). This was later modifed by Sen and Ray (1964), and these
feeding practices was revised by Sen, Ray and Ranjhan (1978) with some research
on Indian breeds. Kearl in 1982 compiled data on nutritional requirements of dif-
ferent livestock species and values of different feedstuffs from various developing
countries in a systematic manner. After reviewing these facts, India compiled these
standards, and as a result, nutritional requirements of cattle and buffalo were pub-
lished by ICAR in 1985 and 1998. After this publication, a change in productivity
of Indian livestock and nutritive values were experienced in new verity of crop, and
modern agriculture practices were introduced, which allows new methodologies for
animal nutrition. In 2013, ICAR came out with nutrient requirement of cattle and
buffaloes and other species to enhance their productivity. Unfortunately, feeding
standards have not been popularized to all dairy farmers due to lack of aware-
ness among farmers as well as lack of resources. In fact, there is a need to provide
knowledge of feeding practice to farmers; though there are many government bod-
ies working on these issues in rural areas, there is still need to enhance the overall
productivity. Therefore, in present research, an effort has been made by to provide
a least cost balanced ration to livestock at different conditions.
1.3 Essential Nutrients

Water is one of the most important nutrients for life. For cattle and buffalo an ade-
quate amount of water is necessary according to body weight, but in general cattle
and buffalo need 5.5 to 8.0 and 6.5 to 10.0 litres of water for every 100 kg of body
weight per day depending on atmospheric temperature, respectively (Ranjhan,
1998). Water intake may vary according to temperature and dryness of feed; for
example if cattle eat healthy grass in spring, they will drink less water than the
cattle grazing in summer. Feed contains various amounts of moistures; hence,

nutrient contents are expressed in a dry matter (DM) basis to maintain uniformity.
For feeding of dairy cattle and buffalo, dry matter is what needs to be consid-
ered for feeding other nutrients in adequate quantity and to satisfy animal satiety.
Underfeeding as well as overfeeding animals needs to be avoided, as underfeeding
causes less production and may lead to nutritional disorder, whereas overfeed-
ing can cause high cost and reproduction issues and excess nutrient excretion,
which cause environmental pollution. DM intake (DMI) depends upon body weight
of cattle and buffalo, where body weight can be predicted through body measure-
ment (ICAR-NIANP, 2013). Several factors affect DMI including body weight,
milk yield, milk fat percentage, environment temperature, density of nutrient in
feed, water intake and overall health status, etc., whereas low DMI may lead to
nutritional stress and cause less production, infertility and health deterioration.
1.3.1 Energy
Per se, energy is not a nutrient, but it is an important component for all types of
body functions which needs to be supplied through diets because dairy cattle and
buffalos need to maintain their energy level for body maintenance, reproduction,
lactation and growth and during pregnancy. Energy requirement for dairy animal
are expressed as total digestible nutrients (TDN), where the requirement of TDN
will differ according to animal physiological condition.
1.3.2 Protein
Protein is basically made up of amino acids by peptide bonds, which are used
to make tissues. Protein is important for growth and milk production and to
maintain body conditions. Livestock use amino acids from digested protein to
repair and to build tissues and also utilize microbial protein produced in rumen
during the protein degradation process. Because of rumen microbes in cattle,
rumen can break down about 80% of protein in the feed and form microbial
protein. Protein requirements for dairy animals usually are expressed as crude
protein (CP).
1.3.3 Minerals
Minerals are important for various body functions of dairy animals. Minerals,
along with proteins, form bone and teeth, and some minerals help in nerve
impulse transmission and to carry oxygen. These are basically two types – macro
and micro. For dairy animals, the major minerals are macro minerals such as
calcium, phosphorus, magnesium, sodium, potassium, chloride and sulphur
which is required in diet at a higher level, which is measured as a percentage of
dry matter. Based on different physiological conditions of livestock, minerals
must be included for optimum performance and health. Therefore, by consider-
ing these facts, the required value of DMI, TDN and CP for dairy animals has
been calculated as per the ICAR 2013 standards, while values of calcium and
phosphorus are calculated according to National Research Council (NRC) 2001

standards. These values are fxed by ICAR and NRC, but a small amount of
variation has been incorporated after discussion with qualifed animal nutrition-
ists from ICAR-National Institute of Animal Nutrition and Physiology (NIANP),
in Adugodi, Bangalore.
1.4 Classification of Feedstuffs

Feed is classifed as concentrates and roughage (dry and green) for dairy animals.
Common green roughages are cereal grasses, which may be annual or perennial
grasses. There are two types of green roughage, cereals and leguminous. Legumes
have advantage over cereal grass due to their nitrogen fxing, which means that
these plants contain more protein than cereal grass because they can fx atmo-
spheric nitrogen (with the help of bacteria), and they are not dependent on nitrogen
content present in the soil. Dry roughages include straw from rice, barley wheat
and ragi, and stovers from sorghum, maize, Bajra etc., are widely used as dry
roughage in feed. Roughage is feed which have high fber content, usually more
than 18% on DM basis. The concentrate contains high amount of nutrients with
less fber usually less than 18% on DM basis. The most common concentrate is
cereal grains such as maize grain, sorghum grain, broken rice, wheat etc., as well
as oil seed cake of various kinds and milling by-products. These concentrates are
very rich in energy compared to roughage. Concentrates have a high amount of
energy (TDN) and less fber (less than 18% of DM); hence it is denser than rough-
ages. Nutrient requirement of a dairy cow/ buffalo is calculated based on the nutri-
ent requirement prescribed by ICAR-NIANP (2013). These requirements include
DM (kg/d), TDN (kg/d), CP (g/d), Ca (g/d) and P (g/d). The nutrient requirements
of a cow/buffalo are met by feeding the required proportion of concentrate and
green and dry roughages. This required proportion needs to be calculated, and the
fnal ration (amount of feed eaten by cow for a day) is prepared by mixing green
and dry roughages and concentrate together. The mixture of concentrate and green
and dry fodder is called the total mixed ration. While preparing feed, all the nutri-
ents (DM (kg/d), TDN (kg/d), CP (g/d), Ca (g/d) and P (g/d)) must be balanced,
and at the same time cost of the fnal feed should be minimized. Preparation of
least cost feed formulation is a challenging task for any nutritionist, and appli-
cation of optimization techniques of operational research will provide solutions
for this kind of problem. A goal-seeking program of operational research will
be a useful tool to formulate rations for dairy animals. In this program, required
nutrients (DM (Kg/d), TDN (Kg/d), CP (g/d), Ca (g/d) and P (g/d)) of a dairy for
physiological state (yielding 10 litre/d milk production with 4% fat and pregnancy
in third trimester) are fxed as constraints, and the amount of feed required in
terms of concentrate and green and dry fodder is calculated by applying heuristic
techniques (genetic algorithm). Some of the common feedstuffs commonly used
for ration formulation with its nutrient values are listed in Table 1.1. While formu-
lating least cost balanced ration, the selection of feedstuffs is depending upon the
availability in the respective region.
TABLE 1.1
Nutrient Content of Common Feeds and Fodder for Dairy Cattle and Buffalo on DM
Basis
Name of the ingredients DM % CP % TDN % Ca % P%
Green roughage
Bajra Napier CO-4 grass 25 8 52 0.144 0.09
Maize fodder 25 8 60 0.53 0.14
Sorghum Co-FS 29 fodder 90 7 50 0.12 0.09
Berseem 90 5 45 0.65 0.05
Dry roughage
Paddy straw 90 5.13 40 0.18 0.08
Maize stover 90 3 42 0.145 0.027
Ragi straw 90 6 42 0.15 0.09
Concentrate
Maize 90 8.1 79.2 0.018 0.27
Soya DOC 90 42 70 0.018 0.225
Copra DOC 90 22 70 0.036 0.9
Cotton DOC 90 32 70 0.036 0.9
Wheat bran 75 12 70 1.067 0.093
Gram chunnies 90 17 60 0.108 0.234
Cotton seed 90 16 110 0.3 0.62
Chickpea husk 90 18 45 0.3 0.62
Concentrate mix Type I 90 22 70 0.5 0.45
Concentrate mix Type II 90 20 65 0.5 0.4
Minerals
Calcite 97 0 0 36 0
Grit 96 0 0 36 0
MM 90 0 0 32 6
DCP 90 0 0 24 16
Soda bicarbonate 90 0 0 0 0
Salt 90 0 0 0 0
TM mix 98 0 0 0 0
Urea 95 287.5 0 0 0
1.5 Application Area

According to past surveys, it was clear that farmers are not feeding dairy cattle
according to the nutrient requirement suggested by the scientifc community due to
unawareness and unavailability of proper feedstuffs. Animal feed plays a large role
in the sustainability performance of animal production systems, and the choice of
diet affects animal production. Diet formulation could be a method of blending
totally different feedstuffs in appropriate proportions so that animals can consume
it, and it provides required nutrients to cattle.
In India many livestock organizations have reported shortages of feedstuff;

therefore, Karnataka dairy farmers faced many problems in increasing the perfor-
mance of livestock due to the inadequacy of feedstuff in both quality and quantity.
In addition to that, many feed resources that could have a major impact on livestock
production are either unused or poorly utilized by the farmers, and a critical fac-
tor in this regard is lack of awareness about feedstuffs available in their respective
regions and nutritional values. Even though in the past many decades a reasonable
growth has been seen in dairy animals, it is necessary to increase their productivity
(Ravinder, 2017). Milk production is one of the major contributors to the growth of
the Indian economy, so it is necessary to formulate a low-cost, balanced ration for
livestock. To formulate the ration, it is necessary to have the best possible informa-
tion about feedstuffs, nutrients contained inside the feedstuffs and, additionally,
which sort of dairy animal should be supplied with such a ration to guarantee an
ideal production at a cheaper rate. Various categories of dairy cows and buffaloes
have diverse prerequisites for vitality (sugars and fats), proteins, minerals and vita-
mins to keep up their various functions, such as body repair, proliferation, milk
production and fetus growth during the third trimester of pregnancy. The formu-
lated ration should be rich in energy, protein, phosphorus, minerals and vitamins;
in addition, it ought not create any genuine stomach-related problems or unfavor-
able impacts on the dairy cattle (Gupta et al., 2013a). For example, pregnant cattle
at third trimester not only require nutrition for body maintenance but also for milk
production with adequate amount of fat percentage and fetus growth. It also is nec-
essary to avoid the overfeeding of nutrients to cattle (Mudgal et al., 2003), which
may affect milk production and their health. For this reason, an animal nutrition
expert is required for information about nutrition requirements.
2
Review of Literature
CONTENTS
2.1 Introduction .................................................................................................... 9
2.2 Linear Programming in Ration Formulation ............................................... 10
2.2.1 Linear Programming Based Software............................................. 13
2.2.2 Limitations of Linear Programming in Ration Formulation .......... 13
2.2.3 Overcoming the Limitations of Linear Programming .................... 14
2.3 Nonlinear Programming .............................................................................. 14
2.4 Goal Programming ...................................................................................... 18
2.5 Genetic Algorithm ....................................................................................... 19
2.1 Introduction
Feed formulation is the process of quantifying the amounts of different feed
ingredients that need to be put together to form a single uniform mixture (diet)
for animals that supplies all their nutrient requirements. It is concerned with the
allocation of nutrient ingredients to the animal in terms of yield and weight gain at
the least possible cost. It is one of the central operations of the animal industry, as
the performance and the optimal growth of animals is entirely dependent on their
feed intake. To meet the animals’ requirement at a particular stage of production,
it is very important to formulate the diet effciently with low cost, as feed costs
account for 40–80% of the total costs in animal production (Pond et al., 1995),
75–85% in pig production (Gallenti, 1997: 15–19) and over 60% in poultry pro-
duction (Rose, 1997), and in milk production feed costs are the largest expense
(Bath, 1985: 1579–1584). Therefore, procedures that reduce feed costs are likely
to increase net incomes in agriculture. Feed formulation is one of the areas that
one can use to reduce the cost of feed (Nabasirye et al., 2011: 221–226). Rations
for livestock should be formulated in such a way that they should supply all essen-
tial nutrients and energy to maintain the vital physiological functions of growth,
reproduction and health of animals. To formulate the low-cost diet, planning the
linear programming-based diet formulation has come into existence with the help
of mathematicians. This method is widely used in commercial dairy farms and
commercially available software. The least cost formulation of rations based on
linear programming optimizes the combination of feed ingredients which supplies
the required amount of nutrients at least cost (Rossi, 2004). The major nutrients to
DOI: 10.1201/9781003231714-2 9
be supplied in a diet are energy, protein and minerals such as calcium and phos-
phorus for maintenance, growth, pregnancy and milk production of dairy animals.
These nutrients should be supplied in required quantity to meet the specifc per-
formance targets. These entire concepts need a professional knowledge of animal
nutrition to consider the requirements of the nutrients and animals’ capability to
digest and then assimilate the nutrients from various available ingredients.
Livestock ration formulation models are developed for commercial purposes as
well as for development, using various mathematical techniques for several decades.
Some of them use the Pearson’s square method: the Pearson’s square ration formu-
lation procedure is designed for simple rations and has been used for many years.
It is direct and easy to follow and is useful in balancing protein requirements. The
limitation of this technique is that it satisfes only one nutrient requirement and uses
only two feed ingredients. It therefore has limited application where farmers must
formulate diets balanced for protein, energy, minerals and least cost (Chakeredza
et al., 2008: 2925–2933). The simultaneous equation method is one more method
that uses a simple algebraic equation. The advantage of this over the Pearson’s
square method is that it is useful in considering more than two feed ingredients at
a time when balancing rations is more complex. The two by two matrix method
solves two nutrient requirements using two feed ingredients. A 2 × 2 matrix is set
and a series of equations are done to come up with the solution to the problem.
Trial and error method: This method is used for formulating rations for swine
and poultry. As the name implies, the formulation is manipulated until the nutri-
ent requirements of the animal are reached. This method makes possible a diet
formulation that meets all the requirements of the animal. The major limitation
of this method is that it is laborious and time consuming before one approaches
a better solution. Many mathematical programming techniques are being used to
formulate rations, such as linear programming, multiple-objective programming,
goal programming, separable programming, quadratic programming, nonlinear
programming and genetic algorithm. In the present work, we are discussing linear
and nonlinear programming techniques in animal diet formulation.
2.2 Linear Programming in Ration Formulation

The history of linear programming (LP) goes back to 1947, when George Dantzig and
his team were working in the U.S. Air Force and found that many military program-
ming and planning problems can be formulated to minimize/maximize a linear form
of proft/cost function whose variables were restricted to values satisfying a system of
linear constraints. Linear programming has its application in various decision-making
processes such as human diet, portfolio optimization, transportation problems, crew
scheduling airlines etc. In many algorithms for optimization problems, linear pro-
gramming is used to solve a sub-problem where, if the feasible region is a subset of
the non-negative portion defned by linear equations or inequalities, and the objective
function to be minimized or maximized is linear, then we have a linear program-
ming problem or LPP (Meyer, 1985). Selecting the best alternative out of many pos-
sibilities is called optimization. Patrick and Schaible (1980: 417–458) stated that LP
Review of Literature 11
is technically a mathematical procedure for obtaining a value weighting solution to a

set of simultaneous equations. The earliest systematic method for numerically solv-
ing an LPP was developed by Dantzig and is called the simplex method. The simplex
method is an algebraic iterative procedure which exactly solves an LPP problem in a
fnite number of steps or gives an indication that there is no solution or an unbounded
solution. Most of the computational techniques currently in use are based on the sim-
plex approach. Most companies would like to maximize profts or minimize costs
with limited resources; such issues can be characterized as LPP. There are hundreds
of applications of linear programming in agriculture (Taha, 1987). One of the major
applications is in livestock feed formulation. Linear programming is the common
method of least cost feed formulation, and for the last 50 years it has been used as
an effcient technique in ration formulation. It has formed the basis for livestock feed
formulation since Waugh (1951) defned the feeding problem in a mathematical form.
He was not an animal nutritionist by profession but, as he confessed, was trying to
look for something suitable to lighten the animal nutritionist’s job. Since then, LP has
been used widely for diet formulation. While formulating rations by linear program-
ming, frst and foremost, all available raw materials and feed ingredients are selected
which are to be included in ration, and then a set of constraints on feed ingredients is
set up. An objective function is then formulated as per the available feed ingredients
and constraints to achieve the objective. While constructing a linear programming
model, the following assumptions are made:
Linearity: Feed formulation by LP is mainly based on the linearity between

animal yield and nutrient ingredients included in the diet. Linearity
means the amount of each resource used and its contribution to the objec-
tive function is proportional to the value of each decision variable.
Simple objective: There is single objective (can be either maximization of
animal yield or minimization of cost of the diet) which is a mathematical
function.
Additivity: This means that the sum of resources used by different activities
must be equal to the total quantity of resources used by each activity for
all the resources (Dantzig, 1963).
Certainty: While formulating a linear model for animal rations, it is assumed
that all parameters of the model and consumption of resources are known
with certainty.
Divisibility: It is assumed that all inputs into the ration are infnitely divis-
ible. Perfect divisibility of outputs and resources must exist.
Non-negativity: Decision variables cannot be added to the fnal objective
function in a negative way. More precisely, the solution values of decision
variables are allowed to assume continuous and non-negative values.
Finiteness: The constraints and the variables must be fnite so that the ration
can be programmed. Hence, a fnite number of activities and constraints
must be employed (Gale et al., 1951).
Proportionality: This implies that the contribution of each variable to the
objective function is directly proportional to each variable (Al-Deseit, 2009).
All these assumptions indicate that the objective function and all the constraints
must be characterized by a linear relationship among decision variables in the
LP model. LP has been used as a basic tool for animal feed formulation since
1950. Interest in livestock ration formulation accelerated, and a laboratory diet was
considered as an important variable. During the 1970s, the American Institute of
Nutrition, National Academy of Science, Institute of Laboratory Science, Institute
of Laboratory Animal Resources and Laboratory Animal Centre Diets Advisory
Committee supported the use of standard reference diets in biomedical research.
A linear model was developed (Van de Panne and Popp, 1963) to formulate an
optimized composition of cattle feed. In this LP model, the coeffcients of con-
straints can be considered as stochastic. Feed utilization is described in terms of
feed ration, response gain, biological function and economic effciency. Nutritional,
genetic and physiological responses are interpreted as a function of gain and feed
intake (Meyer and Garrett, 1967); the incorporation of information on animal per-
formance into the linear programming derivation of optimum livestock rations is
taken into consideration (Townsley, 1968). Another model was developed using
linear approximation of chance constrained programming (Olson and Swenseth,
1987). The LP model can be solved for a complicated set of nutrient requirements
to give a relatively well-balanced ration (Haar and Black, 1991: 541–556). Sklan
and Dariel (1993) developed a nutritional program for high producing dairy herds
to attain effcient and proftable levels of milk production. The main goal in the
application of LP in feed formulation is the production of least cost rations that
produce satisfactory results. A nutrition program was developed for high producing
dairy herds to attain effcient and proftable satisfactory results. It has been widely
used in livestock rations (Lara, 1993: 321–334), and to formulate feeds, nutrition-
ists should be knowledgeable in diet specifcations as well as result interpretations.
LP is one of the most important techniques to allocate the available feedstuffs in a
least cost broiler ration formulation (Dantzig, 1951; Aletor, 1986; Ali and Leeson,
1995). Olorunfemi et al. (2001) reviewed extensively the use of linear program-
ming in the least cost formulation for aquaculture. Olorunfemi et al. (2001) also
applied linear programming into duckweed utilization in low-cost feed formula-
tion for broiler starter. In 2002, Darmon used LP as a tool to optimize human
nutrition. A model developed by Tedeschi et al. (2004) is used to represent the
effciency of nutrient use in relation to proftability of dairy farms. Guevara in 2004
developed a cost analysis spreadsheet and validated the effcacy on milking and
custom heifer operations. A stochastic linear programming-based Excel workbook
was developed that consists of two worksheets illustrating linear and stochastic
program methods. This Excel sheet was set up to calculate the margin of safety
value according to probability. The multiple-objective programming was applied
for feed formulation with the objective of minimizing the nutrient and minimizing
cost. This study introduced a dual model in linear programming to obtain the price
of resources that take part in optimization. The price of nutrients resource showed
degree of infuence of a diet’s cost by increasing or decreasing the expected nutrient
values. When the price of some kind of resource is zero, it means that reaching this
nutrient value does not have infuence on a special diet least cost within a particu-
lar value (Xiong et al., 2003). Therefore, for the past seven decades, use of linear
programming technique has been used to formulate least cost dairy feed (Waugh,
1951; Rehman and Romero, 1987; Zhang and Roush, 2002; Hadrich et al., 2008;
Zgajnar and Kavcic, 2008; Al-Deseit, 2009; Almasad et al., 2011; Saxena, 2011a).
2.2.1 Linear Programming Based Software

A few software packages are available based on linear programming for feed
formulation such as FEEDLIVE, FEEDMU, WINFEED, ECOMIX, BESTMIX,
OPTIMIX, MIXIT, WINPAS, EGGOPRO, FEEDMANIA etc. FEEDLIVE
software is meant for feed formulation, and it is bilingual for Thai and English.
FEEDMU is simple feed formulation software which is based on the trial and error
method and simplex method of linear programming. FEEDMU is upgraded; the
new version is FEEDMU2, which uses .NET framework 2.0 technology. FEED
FORMULATION is software used to formulate an optimum diet. WINFEED is a
very simple and straightforward software which prepares least cost formula diets
for ruminants and non-ruminants by using linear and stochastic programming.
LINDO (Linear Interactive and Discrete Optimizer), MATLAB, GAMS (General
Algebraic Modeling System) and Microsoft Excel Solver are some of the numer-
ous packages dedicated to solving LP problems (Nabasirye, 2011). There are many
advantages of using a computer for feed formulation. Some of them are:
• It is convenient and saves manpower.

• It eliminates human error both in calculation and in speed.
• It allows least cost ration formulation using specifc information fed into
the computer.
• It is a choice for commercial feed millers who handle a large number of
ingredients.
• Least cost minimizes the cost of rations given a certain set of ingredients
and their nutritional content, which is done in real time using a computer.
(Afolayan, 2008: 1596–1602)
2.2.2 Limitations of Linear Programming in Ration Formulation

However, there are many limitations in using linear programming (LP) while for-
mulating rations, which lies in exclusive reliance of cost. It is very rigid assumption,
as more often the decision maker is not interested in an economically optimal ration
that achieves a compromise amongst several conficting objectives such as minimi-
zation of cost, imbalance of nutrients supplies and satisfaction of conditions such as
the calcium–phosphorus ratio, roughage–concentrate ratio etc. Another weakness of
LP is the rigidity with which given nutritional requirements must be met (Rehman
and Romero, 1984). Briefy, linear programming is an effective tool of fnding the
least possible cost of a diet that satisfes a set of nutrient specifcations. Although LP
has been used widely in practice, with noticeable success (Black and Hlubik, 1980),
the assumptions considered while formulating the linear model remain its weakness.
Linear programming optimal solutions are based on a linearity assumption which
in practice may not be true. Since varying constraints need to be considered, the
feed mixing problem has become increasingly diffcult to solve using linear pro-
gramming alone. In varying situations, LP is found insuffcient to overcome all the
complexities of the problem. It suffers unnecessarily from over-rigid specifcation
of nutritional and other requirements. Some relaxation of the constraints imposed
would not seriously effect an animal’s well-being and performance.
Over the years, several issues have arisen in feed mixing problems such as ingre-
dient variability. Among them are that the nutrient levels of feed ingredients are
unstable and fuctuate (Munford, 1996; Panne and Popp, 1963), that there can be a
nutrient imbalance in the fnal solution and that price variability means prices of
feed ingredients are not constant, making the solution infeasible. These problems
include the singularity of the objective function and the rigidity of the constraint
set. The singularity of the objective function refers to the reliance on cost alone as
the most important factor in determining the composition of the ration. Lara (1993)
also criticizes practical applications of LP due to the restrictions placed on decision
makers’ preferences through a singular objective function. Producers are likely to
have many objectives in mind while formulating a ration (Tozer and Stokes, 2001).
Mitani and Nakayama (1997: 131–139) pointed out three limitations of linear pro-
gramming model while formulating the ration:
LP models assume nutrients levels are fxed; however, nutrient levels in

feed ingredients are unstable and fuctuating. When the variability among
ingredients is neglected, the probability of meeting nutrient restriction is
only 50% (Pesti and Siela, 1999). It is hard to determine a good balance
of nutrients in the fnal solution of the linear programming method. If
only the minimum levels of nutrient requirements are placed, there is a
probability for nutrient imbalance to arise in the fnal solution. When the
variation is small, the quality of balanced nutrients improves. (Zhang and
Roush, 2002). The constraint is over-rigidity of nutritional specifcation
and requirement, which means no constraint violation is allowed in LP.
This normally leads to an infeasible solution.
2.2.3 Overcoming the Limitations of Linear Programming

Keeping the limitation of LP in mind, various techniques have been proposed in
the feld of ration formulation such as goal programming, multi-objective goal
programming, multi-objective fractional programming, nonlinear programming,
chance constrained programming, quadratic programming and risk formulation.
All these methods have advantages. However, goal programming and the inte-
grated linear goal programming approach and its application in the ration formula-
tion problem will be discussed at length in Chapters 4, 5 and 6.
2.3 Nonlinear Programming

In linear programming models, the characteristic assumption is the linearity of the
objective and constraint functions. Although this assumption holds in many practi-
cal situations, we still come across many situations where the objective function
and/or some or all of the constraints are nonlinear functions. Such problems are
referred as nonlinear programming problems (NLPP). If the objective function and
the constraints are all convex functions (a function which, when plotted, results in
a curve always curving upwards or not curving at all is called a convex function),
the problem is said to be a convex programming problem. The method of solving
an LP problem is based on the property that the optimal solution lies at one or more
extreme points of the feasible region. This limits our search to corner points, and
a fnite solution is obtained after a fnite number of iterations, as in the simplex
method. But this is not true for an NLPP. In such problems, the optimal solution
can be located at any point along the boundaries of the feasible region or even
within the region. And due to nonlinearity of the objective function and constraints,
it becomes diffcult to differentiate between the local and global solution. Every
linear programming problem can be solved by the simplex method, but there is no
single technique which can be claimed to effciently solve each and every nonlinear
optimization problem. In fact, a technique which is effcient for one nonlinear opti-
mization problem may be highly ineffcient for solving another NLPP. A variety
of computational techniques for solving NLPP are available (Himmelblau, 1972;
McCormick, 1970). However, an effcient method for the solution of a general NLPP
is still a subject of research. There is a wide variety of nonlinear programming
problems. Some of the most important types are constrained and unconstrained
NLPP, separable programming, convex programming, non-convex programming,
quadratic programming and so on. In a linear-constrained NLPP, all the constraints
are of a linear type, and the objective function is nonlinear. And in a nonlinear-
constrained NLPP, all the constraints are nonlinear. Many have tried to use uncon-
strained optimization methods for solving nonlinear problems with constraints. A
successful and frequently used approach is to defne an auxiliary unconstrained
problem such that the solution of the unconstrained problem yields the solution
of the constrained problem. Goldfarb has extended the Davidon–Fletcher–Powell
method to handle problems with linear constraints utilizing the concept of gradi-
ent projection (Goldfarb, 1969: 739–764). The method was generalized by Davies
to handle nonlinear constraints (Davies, 1970). Box developed a constrained ver-
sion of the simplex method (Broyden, 1967: 386–381). Another method that uses
the simplex technique in constrained optimization was proposed by Dixon (1973a:
23–32). Another important class of methods used for solving a constrained NLPP
is known as penalty function methods. In these methods the constrained problem
is converted into an unconstrained problem or a sequence of unconstrained prob-
lems in which there is a severe penalty for the violation of the constraints. Another
class of methods for solving an NLPP is the family of exact penalty function meth-
ods. The pioneering work in this feld has been done by McCormick (1983). The
sequential linear programming approach for constrained optimization problems
fnds the optimal solution by repeated linear approximation of a nonlinear prob-
lem and using linear programming techniques to solve it. Attempts have also been
made to obtain the solution of both constrained and unconstrained optimization
problems by solving the Kuhn–Tucker (KT) conditions directly. Methods under this
category are sometimes classifed as multiplier methods. Quasi-Newton type meth-
ods have been used to obtain the solution to the system of equations representing
KT conditions (Powell, 1977). He also tried to obtain solutions to constrained and
unconstrained NLPP by using a least square algorithm for solving the system of
nonlinear equations representing the KT condition of the problem. In the absence
of convexity, the methods discussed here at the most guarantee a local optimal
solution. Keeping in view the practical necessity and the availability of fast com-
puting machines, many computational techniques are being developed to fnd the
global optimal solution. The methods for solving global optimization problems are
broadly classifed as deterministic and probabilistic methods. The deterministic
methods try to guarantee that a neighborhood of the global optima is attained. Such
methods do not use any stochastic techniques but rely on a thorough search of the
feasible domain. They are applicable to a restricted class of functions only, such as
Lipschitz continuous functions. In stochastic or probabilistic methods, two phases
are generally employed. In the frst phase, or the global phase, the function is evalu-
ated at a number of randomly sampled points. In the second phase, also called the
local phase, these points are manipulated by local searches to yield a possible can-
didate for the global minima. These methods are preferred over the deterministic
methods because they are applicable to a wider class of functions. Some of the
earlier methods of this category are pure random search methods by Bremermann
and Anderson (1989) and Brooks (1958: 244–251). Dixon suggested a method in
which any local search method can be used to search the global solution by start-
ing repeatedly from different initial points chosen stochastically. Price presented a
controlled random search technique (RST) method in which the simplex approach
is used on a random sample of points to yield a better point at each iteration and the
method in the limit converges to the point of global minima Dixon (1972a, 1972b,
1973a, 1973b, 1976) and Bharti (1994) have updated the RST method of Price for
solving the constrained NLPP (Dixon 1973a).
Bharti has tried to improve the performance of various controlled random
search algorithms (1994). Linear programming is an effective method of fnding
the least possible cost of a unit that satisfes a set of nutrient specifcations. An
animal’s nutritional requirements hold constant only within a single performance
level. Changes in diet composition may cause variations in feed intake, animal’s
nutritional requirements and animal performance. The traditional LP formulation
ignores animal performance and feed intake. There are multiple least cost diets at
different levels of nutrient concentrations and corresponding animal performances.
Thus, a single LP formulation will not necessarily converge to the minimum cost
of production or maximum proft, both nonlinear functions of diet composition
(Tedeschi et al., 2005). Low-cost balanced ration formulation for livestock involves
omitting ingredients with an inclusion rate below some fxed threshold and round-
ing other ingredients to realistic weighing quantities. It is possible to incorporate
these constraints, but then it will not be possible to solve by linear programming,
as the nonlinear effect of variables will then be included (Saxena, 2011c: 1–5). It
is necessary to control a ratio of nutrients in a formulation – for example, a simple
ratio such as that of calcium to phosphorus, or a more complicated one such as
forage dry matter to concentrated dry matter in a ruminant complete diet. Also
controlling the dry matter percentage as-fed in a diet involving wet feeds is ratio
constraint. Ratio constraints are nonlinear and fall outside the scope of a linear
programming framework (Munford, 2005, 1989).
The primary objective of feed formulation is to generate a formula to meet the

specifcations set for it at the lowest possible cost. The inherent variability in the
nutrient composition of feed ingredients represents a major risk factor in formula-
tion. The risk associated with nutrient variability must be handled diligently by
feed manufacturers. While under-delivery of nutrients is detrimental to feed per-
formance, over-formulating, that is, specifying nutrients more than requirements is
wasteful and expensive. Nott and Combs (1967) suggested that nutrient variability
in the feed formulation process could be managed by a margin of safety for the
nutrients. They recommended an adjustment of the nutrient means by subtract-
ing 0.5 of the nutrient standard deviation. However, nutrient adjustments are not
appropriate for a linear program. Technically, it is assumed that input values such
as nutrient levels, animal requirements and ingredient costs in a linear program are
linear and are known with certainty. Because the variances of nutrients are used in
the formulation, the algorithm is the square of the standard deviation. Therefore,
nutrient variation as a nonlinear input variable violates the assumptions of linearity
and certainty for the LP. Any attempt to adjust a linear program for nutrient vari-
ability results in costly over-formulation of the ration. The most appropriate feed
formulation method is the use of stochastic programming (William et al., 1994). It
is a nonlinear approach to feed formulation, and it is a refnement in providing mar-
gins of safety for feed formulation. Stochastic refers to the variability of nutrients
and the probability of meeting the nutrient requirement. It is therefore a method of
feed formulation that can effectively incorporate nutrient variability into the for-
mulation process to meet animal requirements with a measured level of certainty.
The LP method is applied as an effective method to achieve the least cost of feed.
But most of the time the price of products and the energy density are ignored in
formulating diets.
A nonlinear programming optimization model was developed to maximize the
margin over feed cost for broiler feed formulation. The model identifes the optimal
feed mix that maximizes proft margin. The NLPP method illustrated the effects
of changes in different variables on the optimum energy density, performance and
proftability and was compared with conventional LP (Guevara, 2004: 147–151).
The study done by Guevara suggests that nonlinear programming can be more
useful than conventional linear programming to optimize performance response to
energy density in broiler feed formulation because an energy level does not need
to be set. In this method, optimum metabolizable energy level and performance
were found by using Excel Solver NLP. Eila, Lavvaf and Farahvash showed that
the optimum level of energy could be obtained by a nonlinear model depending
on the income over cost of energy, so that the price of the produced egg mass is
considered as well as the cost of energy density to achieve the maximum amount
of beneft (Eila et al., 2012: 1302–1306). Chavas, Kliebenstein and Crenshaw used
nonlinear programming to fnd the diets and rates of gain for swine when nei-
ther the length of the feeding program nor the market weights were fxed (1985:
636–646). A model of dynamically optimal cattle purchasing, feeding and selling
decisions was constructed and a simple static model developed to closely approxi-
mate dynamically optimal decisions. The models were solved using nonlinear pro-
gramming software, MINOS (Hertzler, 1988: 7–17). Iterative linear programming
is used to solve two NLPPs of animal diet formulation (Tozer, 2000: 443–451). To
overcome the drawback of linear approximation of the objective function for diet
formulation, a mathematical model based on the NLP technique was developed
by Pratiksha Saxena to measure animal performance in terms of milk yield and
weight gain. The result of developed model was compared with the linear model,
and the comparison shows that NLP gives better results for maximization of animal
yield and weight gain and represents simultaneous effects of all variables together
(Saxena, 2011a: 106–108).
2.4 Goal Programming

Goal programming (GP) is like a linear programming model which allows mul-
tiple goals to be satisfed at a time. Multiple goals are sometimes given priority
according to weights to meet the various goals. It is a branch of multi-objective
optimization, which in turn is branch of multi-criteria decision analysis (MCDA),
also known as the multi-criteria decision making (MCDM) process. Hence, goal
programming is an extension of linear or nonlinear programming involving an
objective function with multiple objectives. While developing a goal-programming
model, the decision variables of the model are to be defned frst. Then the manage-
rial goals related to the problems are to be listed and ranked in order of priority.
Since it may be very diffcult to rank these goals on a cardinal scale, an ordinal
ranking is usually applied to each of the goals. It may not always be possible to
fully achieve every goal specifed by the decision maker. Thus, goal programming
is often referred to as a lexicographic procedure in which the various goals are sat-
isfed in order of their relative importance. The general mathematical model of the
goal programming problem (GPP) is as follows:
Minimize (Z) = ∑i∈m (di+ + di−) Subjected to : ∑ j=1 (aij x j + di− − di + ) = gi

n
where, di+, di−, xj ≥ 0 and (i = 1,2. . .m), (j = 1,2. . .n).
Here di+ is the positive deviation variable from overachieving the ith goal; di− is the
negative deviation variable from underachieving the ith goal, and xj is the decision
variables; aij is the decision variable coeffcient. In the weights method, the single
objective function is the weighted sum of the functions representing the goals of the
problem. The weights goal-programming model is of the form:
Minimize (Z) = ∑i∈m {di(Wi+ + Wi−)}, Subjected to : ∑ j =1 (aij x j + di + − di− ) = gi ,

n
where di+, di−, xj ≥ 0 and (i = 1,2. . .m), ( j = 1,2. . .n). Here Wi+ and Wi− are non-
negative constraints and can be real numbers representing the relative weights
assigned within a priority level to the deviational variables. The parameter di+ rep-
resents positive weights that refect the decision maker’s preference regarding the
relative importance of each goal, while di− is the negative weights of the decision
maker’s preference. The determinant of the specifc values of these weights is sub-
jective. The objective of goal programming is to minimize the deviation to fnd the
solution for which the deviation is at minimum. Therefore, to overcome the limita-
tion of LP in diet formulation, a multiple criteria decision-making technique was
introduced by Romero and Rehman (1984). In 1992 Lara and Romero used the con-
cept of Rehman and Romero (1987) and then applied a multiple goal programming
(MGP) model to introduce the relaxation of nutrient constraints. Lara and Romero
(1994) extended their work, as the relaxation of constraints could reduce the ration
cost. They considered nutritional requirements as targets which might or might
not be achieved. This MGP model is solved using a multiple criteria mathematical
programming technique known as the interactive method. The use of this interac-
tive method is to perform a search over a set of feasible diets. The work of Mitani
and Nakayama (1997) is also an early contribution in diet formulation. Lara (1993)
introduced a second objective, maximization of the inclusion in the diet of the
ingredients available on the farm in addition to least cost. This work is introduced
in a nonlinear framework because the second objective is considered in fractional
form. Tozer and Stokes (2001) used multiple objective programming to reduce
nutrient excretion from dairy cows through incorporation of a nutrient excretion
function into the ration formulation framework. To reduce the nutrient excretion
load, rations are formulated to minimize cost and nitrogen and phosphorus excre-
tion using MGP. Jean dit Bailleul et al. (2001) also developed a multiple objective
optimization method to minimize the cost and excess nitrogen in pig diets. Zhang
and Roush (2002) introduced two objectives minimizing the cost and the nutrient
variabilities for protein, methanone and lysine. MGP applies the same concepts as
GP but is different in terms of modelling the objective separately. Therefore, GP
and MGP have an advantage in handling multiple objectives, including nutrient
variability and reducing nutrient imbalance problems. Castrodeaza et al. (2005)
formulated pig rations considering economic and environmental objectives, incor-
porating advanced nutritional concepts in ratio form and by using an interactive
method by multiple objective fractional programming. They considered the cost
of feed, the lysine/energy ratio, deviation with regard to the ideal values of the
percentage content of amino acids in the protein and the amount of phosphorus.
Zgajnar and associates (Zgajnar and Kavcic, 2008, 2009; Zgajnar et al., 2009) also
combined LP and weighted GP to fnd the optimal ration cost with balanced nutri-
tional requirement. One of the primary objectives was to overcome the limitation
of linear programming such as LP rigidity and not satisfying primary constraints.
2.5 Genetic Algorithm

A genetic algorithm (GA) is a search algorithm which is based on principles
inspired by natural genetic populations to evolve solutions to problems (Holland,
1975; Goldberg, 1989). The main concept is to maintain the population of chromo-
somes, which amounts to candidate solution to a complex problem that evolves over
time through a process of competition. Each chromosome in the population has
ftness to determine which chromosome should be used to form new chromosomes
during the process, which is called selection. The new chromosomes are created
using genetic operators such as crossover and mutation [Goldberg (1989), Holland
(1975)]. GAs has great history of success in search problems; the reason for this is
their ability to exploit the information about an initially unknown search space in
order to bias subsequent searches into useful subspaces, that is, their adaptation.
This is the key factor, particularly in large, poorly understood search space prob-
lems where classical and old conventional techniques are inappropriate in offering
a good search space. Although there are many variants of GA, the fundamental
mechanism operates on a population of individuals and consists of three opera-
tions: a) creating initial population, b) evaluation of individual ftness, c) formation
of a gene pool by selection procedure and d) recombining them using crossover and
mutation procedures. In 1998, F. Herrera et al., reviewed the features of real-coded
genetic algorithms in which different models of genetic operators and some mecha-
nisms available for studying the behavior of this type of GA have been compared.
This review stated that initially the use of RCGA appeared in applications such as
chemo metric applications (Lucasius and Kateman, 1989) and use of meta operators
(Davis, 1989). Real coded GA has been used mainly for numerical optimization in
continuous domains. Later, in 1994, Darrell Whitey covered the canonical genetic
algorithm as well as more experimental forms of genetic algorithm, including
parallel island models and parallel cellular genetic algorithms. He also illustrates
genetic searches by hyperplane sampling and reviewed the theoretical foundation
of genetic algorithms which include the fundamental theorem of a genetic algo-
rithms called a “schema theorem” (Holland, 1975) as well as an exact model of the
canonical genetic algorithm. Therefore, GA has been used for solving nonlinear
programming models as well as multi-objective programming models. It focuses
on the initialization process, evaluation process, selection process, crossover and
mutation for the nonlinear goal programming problem and concluded that GA is
effective for nonlinear GPPs (Zheng et al., 1996; Gen and Cheng, 1997; Deb, 1999;
Kumar et al., 2012). In 2014 a combined LP with GP for feed formulation displayed
the dip in cost as a major advantage of using the goal-programming approach
(Ghosh et al., 2014). Therefore, in view of an exhaustive review of literature, it
is observed that no techniques are suitable for formulating rations in a least cost
manner for dairy cattle; thus, a comparative study was planned to fnd a suitable
technique that is farmer friendly and can be adopted at the farm level.
3
Tools and Techniques
CONTENTS
3.1 Computation of LP Model Using MS-Excel Based Solver .......................... 21
3.1.1 Simplex LP Method in MS-Excel Solver ........................................ 22
3.1.2 GRG Nonlinear in MS-Excel Solver ............................................... 22
3.1.3 Evolutionary Algorithm in MS-Excel Solver .................................. 22
3.1.4 MATLAB (Matrix-Laboratory) ...................................................... 23
3.1.4.1 Inputs for Genetic Algorithm........................................... 24
3.2 Data Handling and Nutrient Bound Calculator ........................................... 26
3.1 Computation of LP Model Using MS-Excel Based Solver

There are mathematical software packages such as Lingo, Mathematica, TORA,
MATLAB, and Excel Solver, etc., which are available widely. Excel Solver, Lingo
and Mathematica are freely available, whereas one needs to use a licensed version
of MATLAB provided by the university. All this software can be used to solve
linear and nonlinear programming problems. The MS-Excel-based spreadsheet
is more popular and widely used because Excel-based spreadsheets provide an
easy environment for data entry as well as data editing. Due to this, one can gain
understanding of how to construct linear models. An MS-Excel based tool known
as “Solver” is available to formulate and solve ration formulation problems. The
importance of this tool was discussed by Macdonald (1995). Solver can be found
in the tool menu of an Excel sheet; if this tool is not found in the tool menu, then it
can be added by clicking on Add-Ins in the tool menu and by checking the Solver
text box. After defning a linear model in a spreadsheet, the model can be solved
directly by Solver, which generates two major reports: the solution report and the
awareness (sensitivity) report. Solution reports give the solution and information/
status of constraints with slack values. An awareness report gives information
about how sensitive the solution is. There are many advantages of using spread-
sheets for ration formulation: it saves time and manpower, it gives less human
error and it is easy to handle more amounts of feed ingredients and constraints
effectively. In MS-Excel 2010, there are three methods available in the Solver tool,
namely, the simplex method, generalized reduced gradient (GRG) nonlinear and
the evolutionary algorithm (EA), which can be used to solve linear or nonlinear
programming models.
DOI: 10.1201/9781003231714-3 21
3.1.1 Simplex LP Method in MS-Excel Solver

The simplex LP method is an iterative method which was developed by George
Dantzig in 1946. According to the Journal of Computing in Science and Technology
(Nash 2000), this method is considered one of the top ten algorithms. This method
can be applied to solve a problem which has a frst-degree equation (linear objec-
tive function) as well as linear constraints. In a plane, the linear function provides a
straight line while plotting. The Excel Solver-based simplex method provides opti-
mal results at the point where two or more constraints intersect, that is, the optimal
result lies at the corner point.
To solve a linear programming problem (LPP) effectively by creating an ideal
starting point, one has to check three basic requirements: a) the problem should be
a maximization problem but can be converted to a minimization problem by rewrit-
ing the objective function with a minus sign; b) linear constraints must be less than
or equal to an inequality; and c) all decision variables should be non-negative. The
simplex LP method is an iterative process which is used to solve LPPs, has frst
degree equations and is applicable if all decision variables as well as constraints are
linear functions. The LPP should have a clear representation of the linear relation-
ship between constraints and variables; that is, if the variables of the LP model are
bounded, then due to the rigidity of the problem, the LP simplex method fails to
satisfy constraints.
3.1.2 GRG Nonlinear in MS-Excel Solver

The GRG method based on the work published by Leon Lasdon in 1973 and Alan
Waren in 1975 is smooth. It deals with an equation involving decision variables or
nonlinear constraints; that is, the derivative of the nonlinear function should not
have any break point, and it should be continuous. If the graph of the function has
a sharp point, it means that the derivative is discontinuous. In MS Excel 2010, the
GRG algorithm picks a starting value from its calculation and hence leads to dif-
ferent answers on each run as it chooses a different starting point every time. The
GRG method provides a global optimum solution if all functions and constraints
are convex. If any function and constraints are non-convex, then it may get stuck
in local optimum solutions. The GRG nonlinear strategy might be utilized to fnd
the solution of any linear problem yet will do so substantially less effectively than
the simplex LP technique. The GRG nonlinear is a proved reliable technique to
solve nonlinear problems, but it can also work for LPPs. The technique takes a long
time and is less effcient for LPPs but is preferable if the linear functions are rigid.
3.1.3 Evolutionary Algorithm in MS-Excel Solver

Evolutionary algorithm (EA) is a term used to portray a computer-based problem-
solving tool which utilizes computational models as an important element to solve
the problem whose functions or constraints are discontinuous and non-smooth.
The MS-Excel-based evolutionary algorithm can obtain optimal solution only if
the problem or model (LP or NLPP) is well scaled. An evolutionary algorithm does
Tools and Techniques 23
not depend on the derivative or gradient information; therefore, it is very diffcult

to determine at intermediate steps whether the result is optimal or not, and Solver
stops with message like, “Solver Converged to Current Solution”. This message
indicates that the ftness of the current population of the trial solution is changing
slowly, which means that Solver has found a global optimal solution. However, less
diversity in the population is a very common problem in evolutionary algorithms,
which indicates that EA fails to return a better solution through mutation, and it
requires another important parameter like crossover to improve the diversity of the
population. Solver also stops when it cannot improve the solution further; that is,
even though EA spends enough (sometimes relatively more) time searching the bet-
ter solution, if it fails to make progress, it will stop with the message “Best Solution
Found”. Therefore, once a solution is found using the Excel-based evolutionary
algorithm, one can use these steps to improve the solution further (just a try): a)
keep the generated solution and restart the solver with an EA option to see whether
it is able to fnd a better answer in less time; b) reduce the convergence value and
increase the maximum sub-problem and the maximum feasible value and restart
the Solver; and c) an increase in population size as well as mutation rate can also
help in fnding the better solution, as an increase in mutation rate will increase the
diversity of the population.
There are different varieties of EA, but MS Excel 2010 uses the basic algorithm,
which starts with an initial random population for evaluating ftness. It only uses
mutation as a parameter to improve the diversity of the population in every genera-
tion. In Chapter 5, the performance of the MS Excel-based Solver is discussed in
detail by solving three different linear programming models of dairy cattle.
3.1.4 MATLAB (Matrix-Laboratory)

MATLAB is a license-based scientifc package product of MathWorks; it is designed
for easy and quick scientifc calculation and graphic visualization in a high level
programming language. MATLAB was originally written by MathWorks scientist
Dr. Cleve Moler for easy matrix software developed in LINPACK and EISPACK
projects. Its frst version was written in the 1970s for course use in linear algebra,
numerical analysis and matrix theory. Therefore, MATLAB is built with a founda-
tion of matrix software where the input element (data) is a matrix which does not
require pre-dimensioning. MATLAB has many inbuilt functions for a variety of
computations; it also has many toolboxes designed for specifc research felds, which
include statistics, optimization, solution of ordinary and partial differential equation
and data analysis. If anyone wants to use MATLAB for the frst time in experiment-
ing with genetic algorithms, a variety of inbuilt functions can be used. Problems can
also be coded in m-fles in a short time using these functions. In addition to this,
a genetic algorithm can also be experimented with in a MATLAB-based toolbox
called an optimization toolbox. The optimization toolbox is a collection of routines
which is written in m-fles, which implement the most useful functions in GA. GA
is a heuristic-based search technique based on natural selection of the fttest, which
is used to solve any sort of constrained as well as unconstrained optimization prob-
lem. GA starts with random number generation and then constantly modifes the
population where, in every step, it randomly selects individuals from the present
population and uses them as a parent to create new children for the next generation.
After many generations the set of the population tends toward an optimal solution.
There are many options and inbuilt functions in MATLAB to modify the param-
eters of the genetic algorithm according to the nature of the optimization problem.
Genetic algorithm differs from old classical methods because old classical methods
create only one point at every iterative step where a string of points approach the
best-ft solution, whereas genetic algorithm creates a population at every iteration,
and the best-ft individual approaches the optimal solution. In MATLAB using an
inbuilt function, in general, genetic algorithm can be defned as follows:
3.1.4.1 Inputs for Genetic Algorithm

Number of Variables = []; Population Size = [];
Linear Inequalities of the form Ax=b
A = []; b = [];
Lower and Upper Bounds
Lower bound = []; Upper bound = [];
ObjectiveFunction = @ {specify the m-fle of objective function};
Setting Options for
Genetic Algorithm
options = gaoptimset.
options = gaoptimset (options,’PopulationType’,’ {specify the function
name}’);
options = gaoptimset (options,’CreationFcn’, {specify the function name});
options = gaoptimset (options,’PopulationSize’, [specify the value]);
options = gaoptimset (options,’EliteCount’, [specify the value]);
options = gaoptimset (options,’TolFun’,[specify the value]);
options = gaoptimset (options,’CrossoverFraction’,[specify the value]);
options = gaoptimset (options,’PopInitRange’,[specify the range]);
options = gaoptimset (options,’Generations’,[specify the value]);
options = gaoptimset (options,’StallGenLimit’,[specify the value]);
options = gaoptimset (options,’StallTimeLimit’,[specify the value]);
options = gaoptimset (options,’CrossoverFcn’, {specify the function name});
options = gaoptimset (options,’MutationFcn’, { pecify the function name});
options = gaoptimset (options,’SelectionFcn’, {specify the function name});
options = gaoptimset (options,’Display’, ‘{specify the function name}’);
options = gaoptimset (options,’Display’, ‘{specify the function name}’);
options = gaoptimset (options,’PlotFcns’, specify the function name });
[x, fval] = ga (ObjectiveFunction, numberOfVariables, A, b,[],[],lb,ub,
[],options);
(https://in.mathworks.com/help/gads/genetic-algorithm/2016/.html)
Gaoptimset provides the options for GA parameters. Populace type represents the
type of data in the population, and by default it takes “double vector”. Creating an
initial population is very important for the algorithm, as the entire population set is
Tools and Techniques 25
based on initial population; hence, the creation function creates the initial popula-
tion in which “gacreationlinearfeasible” is the best option to create the initial popu-
lation which satisfes all bounds as well as linear constraints. Therefore, we have
created an initial population in the range of lower and upper bounds. The popula-
tion size describes how many individuals should be there in each generation. Even
though giving a large population size makes the algorithm run slowly, it is better to
keep a large population size, as by keeping a large population size we are allowing
an algorithm to search in a broader space and there is less chance for the algo-
rithm to stop at local optima. Elite count describes the count of individuals that are
assured to remain in every genesis; it takes only positive integers, or if the value is
not assigned, it takes the default value as a ceiling (0.05 × population size). Elitism
is very important in respect to convergence, as by allowing at least two or three
individuals to survive at every generation, it makes the algorithm converge faster.
Crossover is a very important parameter of the genetic algorithm as it combines two
best-ft parents to create new offspring at every generation. “CrossoverFcn” options
defne the crossover type. There exist various kinds of crossover, such as 1-point
crossover, 2-point crossover etc., but for linear constraints and bounds, “crossover-
heuristic” is the best option for creating offspring. “Crossoverheuristic” returns
offspring which lie on the line that involves two parents; one can specify how far
the children is from the better parent by the parameter ratio. MATLAB takes the
default value of this ratio as 1:2. In general, let P1 and P2 be the two parents and
P1 has high quality robust value, crossoverheuristic will return the offspring by
the formula P2 + ratio × (P1 − P2). The crossover fraction gives the division of each
populace children that are other than world class children (elite children) which
are comprised of crossover children. If the value of the crossover fraction is 1, it
reveals that the algorithm is not having mutation, whereas if the crossover frac-
tion is 0 it refects that algorithm is running without crossover. Therefore, it is a
good choice to keep the value of crossover fraction in between 0.6–0.8 for better
diversity. Like crossover, mutation is also an important parameter of genetic algo-
rithm to maintain the diversity of population; it allows the algorithm to search in a
broader space. Mutation generally help the algorithm make small adjustments in an
entity to create a mutation child. One can defne the mutation option by the function
name “MutationFcn”. Due to rigidity in linear constraints, preference is given to
“mutationadaptfeasible”, where an adaptive feasible mutation generates a direction
which is fexible compared to a previous lucrative or doomed genesis. This muta-
tion chooses a direction and step length that satisfes both linear constraints as well
as bounds. The selection operator helps the algorithm to select best-ft parents for
the next generation; one can specify this option with the function name “selection-
Fcn”. There are many types of selection procedures, such as roulette wheel selec-
tion, uniform selection etc., but preference is given to tournament selection, as the
algorithm will choose each parent by its tournament size 2 at random, and then the
best individual out of these will be selected as parent. Once these parameters are
defned, the genetic algorithm Solver is called to run the program. These param-
eters can be tested on different types of optimization problems (linear or nonlin-
ear), but these parameters need to be modifed as per the requirements. In Chapters
5 and 6, the performance of this MATLAB-based genetic algorithm is tested on
linear and nonlinear goal programming problems for least cost ration formulation
(Kuntal et al., 2016).
3.2 Data Handling and Nutrient Bound Calculator

To formulate the ration problem, the following steps need to be followed: calcula-
tion of nutrient requirement, selection of ingredients, fxing the constraints and
fnding the result. These steps are discussed in detail in Chapters 5 and 6, where the
calculation of the correct nutrients requirement for the specifc body condition of
cattle is frst priority. The nutrient requirement of Indian dairy cattle and buffaloes
is calculated as per the Indian Council of Agricultural Research (ICAR-NIANP,
2013) and National Research Council (NRC, 2001) standards. As discussed in
Chapter 1, since 1957 much research has been done on calculating the nutritive
value of Indian cattle. After many successive experiments, Kearl compiled the
entire data on nutritional requirements of different livestock species, from which
India come across a publication of nutrient requirement of cattle and buffalo in
1985. Since then many corrections have been done in the data, and fnally ICAR
came with another publication of nutrient requirements of cattle and buffalo in
2013. Therefore, to calculate the nutrient requirements, the primary data for the
Mandya district of Karnataka is collected from (ICAR-NIANP), which is located
in Adugodi, Bangalore.
An Excel-based nutrient bound calculator software for cattle and buffalo has
been developed with the help of the Pr. Scientist of NIANP which is based on pub-
lication of nutrient requirements of cattle and buffalo (ICAR-NIANP, 2013) as well
as NRC 2001 standards. The above-mentioned publication has all types of informa-
tion which can be used for nutrient calculation as per the requirement.
4
Binary Coded Genetic Algorithm to
Solve Ration Formulation Problem
CONTENTS
4.1 Introduction .................................................................................................. 27
4.2 Linear and Nonlinear Programming Model for Animal Feed
Formulation (Test-Problem-1) ...................................................................... 28
4.2.1 Linear Programming Formulation .................................................. 28
4.2.2 Nonlinear Programming Formulation............................................. 29
4.2.3 Extended Work Using Genetic Algorithms ..................................... 29
4.2.4 Solution of Nonlinear Programming Model by Genetic
Algorithm ........................................................................................ 30
4.2.5 Graphical Results ............................................................................ 32
4.3 Goal-Programming Model (Test Problem-2) ............................................... 33
4.3.1 Formulation of the Nonlinear Goal Programming Problem ........... 33
4.3.2 Graphical Results ............................................................................ 39
4.3.3 Conclusion ....................................................................................... 42
4.1 Introduction
The farmers of India hold a small number of dairy animals, in which cattle and
buffalo are the main animals used for milk production. With the increase in the
number of high milk producing dairy animals in different part of India due to many
cross-breeding programs, there is a shortage of feed ingredients for dairy animals,
due to which farmers are more conscious about scientifc animal feeding prac-
tices. In addition to that, because of environmental hazards and global warming
problems due to animal excreta, balancing nutrition production for proper health
and excretion is required. Many qualifed animal nutritionists are also fnding dif-
fculty in formulating the low-cost balanced ration for growing dairy animals at
different body conditions. Even though linear programming is a good method for
formulating least cost diet, it has many limitations, as the linear programming tech-
nique requires the objective function to be single and constraints to be rigid (RHS).
Recently, many researchers have described the benefts of the goal programming
method for ration formulation in which the weighted sum goal programming with
a penalty function is used to solve the problem. In this research, an attempt is made
to test the effectiveness of a heuristic algorithm such as a genetic algorithm to opti-
mize the ration formulation on nonlinear and goal programming models.
DOI: 10.1201/9781003231714-4 27
4.2 Linear and Nonlinear Programming Model for

Animal Feed Formulation (Test-Problem-1)
This research is primarily based on the secondary records of the linear and non-
linear mathematical model of livestock ration formulation developed by Pratiksha
Sexena (2011). This nonlinear model was developed for Sahiwal cows during the
second to ffth lactation period, based on experimental data collected from National
Dairy Research Institute (NDRI). Research was carried out on Sahiwal cows during
second to ffth lactation period. It was divided into four categories that are replaced
by four times in Latin-square change-over style. Every lactation period has a dura-
tion of 40 days. These divided categories, namely A, B, C and D, have been sup-
plied with isonitrogenous as well as isocaloric concentrate mixtures which include
groundnut and cotton seed cake individually (see Table 4.1). To fulfll the require-
ment of maintaining 50 grams of dicalcium phosphate (DCP), half of the quan-
tity can be met through a concentrate mixture. A green fodder is added in order to
provide the remaining DCP and to fulfll the dry matter and energy requirements.
Other different concentrate mixtures which contain groundnut, cotton seed cake
(undecorticated and decorticated) were also examined for crude protein, crude fber,
ether extract, organic waste, nitrogen free extract and total ash (Gupta et al., 2013b).
This research work was based on the objective of maximizing the milk yield. Milk
yields and production for which the nutrients are used predominantly rely upon
three variables which might be utilized to amplify production (Gupta et al., 2013b).
Keeping every one of these actualities, milk yield of an animal relies on digestible
crude protein and total digestible nutrients.
4.2.1 Linear Programming Formulation

By using least square method, Pratiksha Sexena has developed a linear relation-
ship between milk yield of cows and nutrient ingredients such as crude protein
and TDN, which describes the weightage of variables x1, x2 and x3. By using this
TABLE 4.1
Composition of Concentrate Mixtures in Respect of DCP and TDN
Control group Cotton seed Cotton seed cake Cotton seed cake
Ingredients (groundnut cake) (whole) (undecorticated) (decorticated)
Groundnut cake 22 10 0 0
Cotton seed 0 57 0 0
(undecorticated) 0 0 44 0
(decorticated) 0 0 0 27
Wheat bran 75 30 53 70
Common salt 2 2 2 2
Mineral mixture 1 1 1 1
Binary Coded Genetic Algorithm 29
relationship, an objective function is formulated where the constraints are applied

according to feeding standards of the National Research Council (NRC, 1981). The
LP model is as follows:
Maximize y = 0.00403908701533 x1 + 0.25469485541324x2

+ 0.02110699233 x3 −8.67895696598672
6
Subject to: x1 = (608.6718, 782.978), x2 = (60.41, 75.943), x3
= (366.0412, 508.9343)
This LP model is solved by the simplex method, which leads to the solution as x1 =
782.97800, x2 = 75.943, x3 = 508.9343 gm/kg metabolic body weight, with maxi-
mum objective value of 24.55.
4.2.2 Nonlinear Programming Formulation

Pratiksha Sexena further developed a relationship between y and x1, y and x2, y and x3 by
using the least square method and by using different degrees (F-test) to fnd the relation
of best ft. Hence, a nonlinear objective function is formulated mathematically with the
relation of variables as per their weightage on milk production, where the constraints are
incorporated as per NRC (1981) standards. The nonlinear model is as follows:
y = 4.1792442x22 − 4.082239204 *10−6 x32 + 0.114836671x1 − 560.07866554x2

+4.145857585 *10−3 x3 +19255.68675. − (1)
Subject to:
x1 = (608.6718,782.978), x2 = (60.641,75.943), x3 = (366.0412,508.9343)
This nonlinear model is solved by using the Kuhn-Tucker method, which gives
the values of the three nutrient ingredients as x1 = (782.97800), x2 = (67.00717),
x3 = (507.79209) gm/kg metabolic body weight, with a maximum objective value of
582.01 gm/kg metabolic weight.
4.2.3 Extended Work Using Genetic Algorithms

In the present study an attempt is made to study the effect of a binary genetic algorithm
over a nonlinear model described by Pratiksha Sexena. The study further extends the
importance of nonlinear livestock ration formulation and fnds its solution by another
heuristic approach called controlled random search technique (RST2, Gupta and
Chandan, 2013). Both the genetic algorithm (GA) and RST2 are heuristic and random
search techniques with the most important features of function evaluation. Briefy,
RST2 was developed by C. Mohan and Shankar (1994) based on random number gen-
eration; it does not consider the mathematical nature of the function and still gives
promising results. RST2 is an iterative process which works on the principle of qua-
dratic approximation. It works in two phases, namely a local as well as a global phase,
without making any prior assumptions about the objective function or constraints. In
the global phase, an objective function is estimated at random based on sample feasible
points, and in the local phase, these feasible points are controlled by local search
options to produce a best ft individual for global minima. GA is an optimization
method which is used to optimize both constrained and unconstrained optimization
problems; it is an adaptive heuristic algorithm which works on Darwin’s principle of
survival of the fttest. The idea behind genetic algorithm is to simulate the process in
the natural system which is necessary for evolution. GA not only provides an alterna-
tive method but also outperforms the other traditional methods consistently. GA starts
with set of solutions (chromosomes) called the population. Solution from one set of
population has been taken to create a new set of population with the hope that the
new set of population is better than the old one. This new set of solutions (offspring)
will be chosen from the population based on its best ftness value. The robust value
is allocated to every individual, where computation of robust value depends upon the
application. At every iteration (generations), candidates are chosen from the set of
randomly generated population for reproduction of new offspring by crossover and
mutation, where crossover will be done with high probability compared to mutation.
The selection of individuals will be depending upon high ftness, but the mean ftness
of population likes to improve the solution from generation to generation. Due to its
random nature, GA improves the chance of fnding the global optimum solution. We
have slightly modifed the technique to suit our requirements and solved the nonlinear
programming problem described by Pratiksha Sexena.
4.2.4 Solution of Nonlinear Programming

Model by Genetic Algorithm
A result of the nonlinear model solved by GA for original bounds Test Problem-1)
is given in Table 4.2, from which the set of solutions obtained are: x1 = (721.3220,
747.1494), x2 = (71.5456, 74.6055), x3 = (432.318, 474.6072), with the global maxima
of 819.1805. Sometimes due to random numbers generated initially and huge inter-
vals, solutions seem to be stuck in the local optimum. Therefore, to avoid this we
have reduced the bounds as follows: x1 = (630.682); x2 = (66.70); x3 = (366.480)
(Test Problem-2). The set of solutions for NLP model (2) by GA for the reduced
bounds are given in Table 4.3. The set of solutions obtained are x1 = (670.5081,
677.1270), x2 = (68.8769, 69.4029), x3 = (428.7671, 463.4859) with the global max-
ima of 593.8254. Also, the comparison result of RST2 and GA is given in Table 4.4.
TABLE 4.2
Solution for NLP Model by GA for Original Bounds (Test Problem-1)
x1 x2 x3
(gm/kg metabolic (gm/kg metabolic (gm/kg metabolic Objective
Iterations weight) weight) weight) function
100 749.0624 73.5973 432.3138 759.5984

200 755.6987 72.4891 495.7266 704.4718
300 744.6752 74.2517 491.5240 796.9501
400 729.9346 72.8592 457.8180 719.0342
500 750.7326 73.6261 475.2268 761.3943
x1 x2 x3
600 747.5276 71.5456 472.9012 664.0140

700 748.0788 72.9356 463.3000 724.8789
800 721.3220 72.9743 465.8431 723.7324
900 759.4326 73.2797 482.3860 743.7354
1000 724.1868 73.1996 484.8339 735.5141
2000 722.4416 72.5642 482.3343 704.1149
3000 747.1494 74.6055 474.6072 819.1805
4000 724.4928 74.5268 472.6285 811.6016
5000 759.6734 73.2089 477.6959 740.0712
TABLE 4.3
Solution for NLP Model by GA for Reduced Bounds (Test Problem-2)
x1 x2 x3
100 672.1468 69.3301 463.4448 591.8256

200 672.0984 68.8769 428.7671 583.8616
300 664.0533 69.3305 439.3848 590.8939
400 662.9170 68.8893 456.7290 583.0166
500 677.1270 69.4029 445.7669 593.8254
600 670.5509 69.4217 444.8161 593.4490
700 662.0822 69.4115 463.4859 592.2789
800 672.0792 69.3436 446.4769 592.0740
900 669.2510 69.1619 460.0039 588.3455
1000 674.4207 69.0258 461.6140 586.5650
2000 674.7802 69.3380 456.8179 592.2804
3000 662.8439 69.0090 453.2494 584.9496
4000 675.4472 69.3782 441.0744 593.1386
5000 670.5081 69.1511 442.2030 588.2859
TABLE 4.4
Comparison of Solution Obtained by GA and RST for NLP Model
RST2 GA
Bounds x1 x2 x3 Obj. func. x1 x2 x3 Obj. func.
Original 608–680 66–68 400–512562–569.9 721.32– 71.54– 432.31– 664–819.18

bounds 747.14 74.60 474.60
Reduced 630–642 66.82– 380–438.5 564–566 670.5– 68.87– 428.76– 583–593.82
bounds 67.22 677.12 69.40 463.48
4.2.5 Graphical Results
Graphical Representation of Maximum Milk Yield Using GA (Original Bounds)

820
X: 3000 GA
800
Y: 819.2
780
Objective function
760
740
720
700
680
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
No of Iteraions
FIGURE 4.1 Maximum Milk Yield Using GA (Original Bounds); Kuntal, IJESET, 2003.
Graphical Representation of Maximum Milk Yield

X: 500
Using GA (Reduced Bounds)
Y: 593.8
594
GA
592
Objective function
590
588
586
584
582
580
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
No of Iteraions
FIGURE 4.2 Maximum Milk Yield Using GA (Reduced Bounds), Kuntal, IJESET, 2003
Graphical Representation of Maximum Milk Yield Using RST and GA (original Bounds)
950 GA
900 X: 3000 RST
Objective function
850 Y: 819.2
800
750
700
650 X: 700
600 Y: 569.9
550
500
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
No of Iteraions
FIGURE 4.3 Maximum Milk Yield Using RST and GA (Original Bounds), Kuntal, IJESET, 2003
Graphical Representation of Maximum Milk Yield

Using RST and GA (Reduced Bounds)
630
GA
620
RST
Objective function
610 X: 500
600 Y: 593.8
590
580 X: 700
570 Y: 566
560
550
540
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
No of Iteraions
FIGURE 4.4 Maximum Milk Yield Using RST and GA (Reduced Bounds), Kuntal, IJESET, 2003
4.3 Goal-Programming Model (Test Problem-2)

This study is extracted from the work of Shrabani Gosh (2011), with an objective
of studying the effectiveness of GA over the goal programming (GP) model of the
ration formulation problem. Briefy, farmers encountered three circumstances for a
fully grown cow with a body weight of 500 kg: a) cow did not produce milk; b) cow
produces various levels of milk with a specifc percentage of fat; c) cow is pregnant
at third trimester, where it requires an extra nutrient supplement.
Therefore, farmers need to feed a low-cost balanced ration to all of these catego-
ries. Based on these three categories, a linear programming model for animal-1, ani-
mal-2 and animal-3 has been developed where animal-1 needs a ration only for body
maintenance, animal-2 needs a ration for body maintenance as well as for 10 litres
of milk production with 4% fat and animal-3 needs a ration for third trimester preg-
nancy. Nutrient requirements as well as the constraints for all three categories were
set as per the Indian Council of Agricultural Research (ICAR) standard (Ranjan,
1998) and as per Mondal et al. (2003). These LP models are solved using the simplex
method. Further, to overcome the limitations of the LP model, a linear goal program-
ming model for three categories of animal has been developed and solved by simplex
method. Since Shrabani formulated the global linear model, in the extension of this
a nonlinear GP model was developed by taking the square root of sum of the squares
of the deviations in which the weight is attached to every goal as per the priority.
4.3.1 Formulation of the Nonlinear Goal Programming Problem

The general GP model is as follows: Min(Z) = ∑i∈mpi(di+ + di−), subjected to
∑
n
(aij x j + di− − di + ) = gi , where di−, di+, xj ≥ 0 and (i = 1,2 . . . m) and (j = 1,2, . . .
j=1
n); di+, di− represents the +ve and −ve deviational variable, which needs to overachieve
and underachieve the ith goal; xj represents the decision variables; aij represents
the coeffcients of the decision variable; pi represents the level of priority which is
assigned to every goal. In this nonlinear GP model with nonlinear objective function
(Gupta et al., 2013b), goals are allotted in such a fashion that all +ve deviational vari-
ables (≤), −ve deviational variables (≥) and both +ve and −ve deviational variables (=)
need to be minimized. The brief description of the goals requirements is as follows:
G-1: dlc−, dlc+ represent an under- and overachievement of least cost ration,
where dlc+ requires minimization.
G-2 and 3: dt−, dt+ represent under- and overachievement of total weight of
feed ingredients, where both require minimization.
G-4: dcp− and dcp+ represent under- and overachievement of the protein
requirement, where dcp− requires minimization.
G-5: dtdn− and dtdn+ represent under- and overachievement of the energy
requirement, in which dtdn− requires minimization.
G-6: dca− and dca+ represent under- and overachievement of the calcium
requirement, where dca− requires minimization.
G-7: dp− and dp+ represent under- and overachievement of the phosphorus
requirement, where dp− requires minimization.
G-8: dg− and dg+ represent under- and overachievement of the grain, maize
and jowar requirement, where dg+ requires minimization.
G-9: db− and db+ represent under- and overachievement of the bran rice and
wheat requirement, in which db+ requires minimization.
G-10: dck− and dck+ represent under- and overachievement of the groundnut cake
as well as the cotton cake requirement, where dck+ requires minimization.
G-11 and 12: dr/c− and dr/c+ give under- and over-achievement of roughage as
well as concentrate ratio (that is, 2:3), where both the deviations require
minimization.
G-13 and 14: dd/g− and dd/g+ represent under- and overachievement of dry-
green ratio (that is, 2:3), where both the deviations need to be minimized.
G-15 and 16: dc/h− and dc/h+ represent the under- and overachievement ratio
of legume and non-legume (that is, cowpea: hybrid Napier in ratio of 1:1),
where both the deviations require minimization. The GP model is nonlinear,
as the objective function involves the square root of the sum of the squares of
the deviations, in which the weight is allocated to every goal as per the prior-
ity (Gupta et al., 2013b). The goal programming model has 12 goals and nine
decision variables, with 24 deviational variables. The objective function has
been formulated by considering deviational variables for each goal (maxi-
mization or minimization) whose target is given on RHS. The nonlinear GP
model for all three levels of animals is given in Table 4.5 where,
1. P1 . . . 16 in the objective function indicates goals set as per priorities, x1 =

Paddy, x2 = Hybrid Napier, x3 = Cowpea, x4 = Maize grain, x5 = Jowar grain,
x6 = Rice bran, x7 = Wheat bran, x8 = Groundnut cake, x9 = Cotton seed cake
2. P (lc, t, cp, TDN, Ca, p, c, b, CK, r/c, d/g, c/h) –, P (lc, t, cp, TDN, Ca, p, c,
b, CK, r/c, d/g, c/h)+ indicates positive and negative deviation
3. x1 . . . 9 ≥ 0, P (lc, t, cp, TDN, Ca, p, c, b, CK, r/c, d/g, c/h) ≥ 0 (Tables 4.6–4.8)
TABLE 4.5
Binary Coded Genetic Algorithm

Nonlinear GP Model with Priority Ranked Goals for Animal at Level-1 to 3 with Objective Function (z) to Different Goals Which Need to Be Minimized
Min(z) = P1d l c +2 + P2 d t −2 + P3d t +2 + P4 d cp +2 + P5d tdn−2 + P6 d ca−2 + P7 d p−2 + P8 d g +2 + P9 d b +2 + P10 d ck +2 + P11d r −2 + P12 d r +2 + P13d d −2 + P14 d d +2 + P15d c −2 + P16 d c +2
c c g g h h
Subjected to constraints Level-1 Level-2 Level-3
Least cost (Rs/kg): ≤7.48 ≤9.01 ≤8.17

4x1 + 2x2 + 4x3 +10x4 + 9x5 +12x6 +12x7 +14x8 + 20x9 + dlc− − dlc +
=1 =1 =1
Total (kg): x1 + x2 + x3 + x4 + x5 + x6 + x7 + x8 + x9 + dt − − dt +
Protein (g/kg): ≥31 ≥108 ≥46.53
30x1 +10x2 +180x3 + 80x4 +110x5 +120 x6 +120x7 + 450x8 + 300x9 + dCP − − dCPP +
TDN/Energy (g/kg): ≥297 ≥693 ≥445

4540 x1 + 550 x2 + 600 x3 + 880 4+ 850 x5 + 660 x6 + 650 x7 + 790 x8 + 790 x9 + dTDN − − dTDN +
Calcium (g/kg): ≥3.8 ≥5.15 ≥3.1

2 x1 + 5.6 x2 +12.8 x3 + 2.7 x4 + 3 x5 + 2.4 x6 + 2.6 x7 + 3.8 x8 + 7.4 x9 + dCa− − dCaa +
Phosphorus (g/kg): ≥2.3 ≥3.78 ≥2.3

1.1 x1 + 3.8 x2 + 5.7 x3 + 4.2 x4 + 3.9 x5 +17.3 x6 +13.7 x7 + 8.4 x8 +13.9 x9 + dP − − dP +
Grain max @ 70% of concentrate (kg): x4 + x5 + dg− − dg+ ≤0.36 ≤0.36 ≤0.36
Bran max @ 50% of concentrate (kg): x6 + x7 + db− − db+ ≤0.30 ≤0.30 ≤0.30
Cake max @ 30% of concentrate (kg): x8 + x9 + dCK− − dCK+ ≤0.21 ≤0.21 ≤0.21
Roughage /concentrate: 3(x1 + x2 + x3 ) − 2( x4 + x5 + x6 + x7 + x8 ) + d r − + =0 =0 =0
− dr
c c
Dry/green roughages: 3x1 − 2(x2 + x3 ) + d d − + =0 =0 =0
− dd
g g
=0 =0 =0
35
− +
Legume/non-legume greens: x2 − x3 + d c − dc
h h
TABLE 4.6
Solution Set of LPP for Level-1, 2 and 3 by GA
36
Generation Levels x1 x2 x3 x4 x5 x6 x7 x8 x9 Z
100 L-1 0.1174 0.0465 0.1002 0 0.2384 0 0.2627 0 0 0.2784
L-2 0.0819 0.0638 0.0813 0 0.3225 0 0.0278 0 0.1 0.0134
L-3 0.0649 0.0396 0.0422 0 0.3142 0 0.2633 0 0 0.5578
200 L-1 0.0678 0.0429 0.0718 0 0.3336 0 0.2201 0 0 0.2861
L-2 0.064 0.0367 0.0959 0 0.3338 0 0.0155 0 0.0623 0.0149
L-3 0.093 0.0381 0.0995 0 0.2960 0 0.2378 0 0.0000 0.586
300 L-1 0.0664 0.0289 0.1258 0 0.3253 0 0.2831 0 0 0.2814
L-2 0.0599 0.0394 0.0516 0 0.3175 0 0.0228 0 0.1155 0.0159
L-3 0.0775 0.0518 0.0857 0 0.2855 0 0.2477 0 0 0.6303
400 L-1 0.0756 0.0327 0.0741 0 0.2387 0 0.2286 0 0 0.2656
L-2 0.0834 0.0509 0.0648 0 0.3389 0 0.0312 0 0.0539 0.0102
L-3 0.0930 0.0381 0.0995 0 0.2960 0 0.2378 0 0 0.5860
500 L-1 0.1039 0.0311 0.0758 0 0.2815 0 0.2605 0 0 0.2852
L-2 0.0534 0.0538 0.1118 0 0.3261 0 0.0194 0 0.0849 0.0136
L-3 0.0378 0.0299 0.1477 0 0.2983 0 0.2648 0 0 0.6098
600 L-1 0.1027 0.0362 0.0943 0 0.3287 0 0.2692 0 0 0.2767
L-2 0.1106 0.0547 0.0655 0 0.3572 0 0.0269 0 0.0957 0.0113
Livestock Ration Formulation

L-3 0.0903 0.0428 0.1197 0 0.3174 0 0.2262 0 0 0.6064
700 L-1 0.0961 0.0289 0.0683 0 0.2651 0 0.2309 0 0 0.2782
L-2 0.0803 0.0507 0.1096 0 0.2991 0 0.017 0 0.1074 0.0119
L-3 0.1154 0.0427 0.0837 0 0.3460 0 0.2506 0 0 0.6247
800 L-1 0.0873 0.0486 0.152 0 0.2806 0 0.2371 0 0 0.2629
L-2 0.1191 0.0524 0.0752 0 0.3464 0 0.0285 0 0.0612 0.0183
L-3 0.0857 0.0379 0.1244 0 0.2960 0 0.2427 0 0 0.5931
900 L-1 0.0667 0.0486 0.0889 0 0.3224 0 0.2772 0 0 0.2864
L-2 0.0700 0.0713 0.0983 0 0.3084 0 0.0182 0 0.0623 0.0116
L-3 0.0311 0.0461 0.0794 0 0.3150 0 0.2284 0 0 0.6406
1000 L-1 0.072 0.1931 0.0000 0 0.2161 0 0.1529 0 0 0.2014
L-2 0.1028 0.0545 0.0876 0 0.3411 0 0.0237 0 0.0230 0.0167
L-3 0.1124 0.0386 0.096 0 0.2928 0 0.2267 0 0 0.4880
TABLE 4.7
Solution Set of the Nonlinear Programming Problem (NLPP) for Level-1, 2 and 3 by GA
Binary Coded Genetic Algorithm

Generations Levels x1 x2 x3 x4 x5 x6 x7 x8 x9 Z
100 L-1 0.0997 0.1663 0 0 0.246 0 0.145 0 0 0.174
L-2 0.1334 0.0589 0.0786 0 0.356 0 0.026 0 0.097 0.022
L-3 0.0507 0.035 0.0926 0 0.342 0 0.262 0 0 0.528
200 L-1 0.0555 0.1887 0 0 0.253 0 0.132 0 0 0.164
L-2 0.0641 0.0709 0.1129 0 0.352 0 0.024 0 0.089 0.027
L-3 0.0601 0.0415 0.1002 0 0.277 0 0.265 0 0 0.499
300 L-1 0.0937 0.1617 0 0 0.231 0 0.197 0 0 0.166
L-2 0.063 0.0688 0.1222 0 0.352 0 0.018 0 0.076 0.020
L-3 0.0872 0.0416 0.0825 0 0.261 0 0.248 0 0 0.466
400 L-1 0.0993 0.2019 0 0 0.215 0 0.134 0 0 0.169
L-2 0.1012 0.0544 0.0499 0 0.295 0 0.026 0 0.069 0.027
L-3 0.0859 0.0444 0.0466 0 0.326 0 0.249 0 0 0.511
500 L-1 0.1078 0.1863 0 0 0.217 0 0.177 0 0 0.168
L-2 0.0634 0.0685 0.0807 0 0.285 0 0.019 0 0.037 0.025
L-3 0.1182 0.0428 0.0476 0 0.256 0 0.257 0 0 0.538
600 L-1 0.0926 0.1887 0 0 0.278 0 0.193 0 0 0.176
L-2 0.1043 0.0809 0.1079 0 0.329 0 0.029 0 0.102 0.028
L-3 0.1144 0.0463 0.0846 0 0.341 0 0.268 0 0 0.535
700 L-1 0.0677 0.1597 0 0 0.219 0 0.200 0 0 0.199
L-2 0.0551 0.0469 0.113 0 0.335 0 0.028 0 0.071 0.023
L-3 0.0821 0.0517 0.1025 0 0.252 0 0.263 0 0 0.556
800 L-1 0.0360 0.1712 0 0 0.227 0 0.161 0 0 0.187
L-2 0.1227 0.0626 0.1185 0 0.374 0 0.026 0 0.086 0.029
L-3 0.087 0.0303 0.0938 0 0.312 0 0.239 0 0 0.563
900 L-1 0.1273 0.1365 0 0 0.194 0 0.179 0 0 0.178
L-2 0.1248 0.0746 0.1132 0 0.346 0 0.018 0 0.054 0.027
L-3 0.0768 0.0368 0.0492 0 0.328 0 0.250 0 0 0.532
1000 L-1 0.1100 0.1518 0 0 0.256 0 0.203 0 0 0.179
37
L-2 0.0531 0.0553 0.0805 0 0.254 0 0.025 0 0.041 0.027
L-3 0.0903 0.0328 0.0863 0 0.289 0 0.273 0 0 0.552
38
TABLE 4.8
Comparison of Decision Variables by RST and GA
GA RST
Variable L-1 L-2 L-3 L-1 L-2 L-3
x1 0.036–0.127 0.0531–0.1334 0.0507–0.1182 0.1–0.196 0.102–0.146 0.102–0.158

x2 0.137–0.2019 0.0469–0.0809 0.0303–0.0517 0.101–0.216 0.027–0.068 0.055–0.077
x3 0–0 0.0499–0.1185 0.0466–0.1025 0.017–0.084 0.141–0.153 0.150–0.187
x4 0–0 0–0 0–0 0.052–0.088 0.062–0.076 0.055–0.098
x5 0.1936–0.2559 0.2536–0.3737 0.2517–0.3421 0.309–0.386 0.314–0.389 0.301–0.395
x6 0–0 0–0 0–0 0.050–0.068 0.059–0.086 0.057–0.095
Livestock Ration Formulation

x7 0.1319–0.2026 0.0177–0.0286 0.2727–0.2685 0.204–0.318 0.012–0.047 0.201–0.299
x8 0–0 0–0 0–0 0.059–0.089 0.050–0.087 0.050–0.088
x9 0–0 0.0409–0.0964 0–0 0.050–0.088 0.101–0.153 0.50–0.649
4.3.2 Graphical Results

Linear Programming graph of level 1 Goal 1
Goal 2
Goal 3
0.5
Goal 4
Deviations
Goal 5
0 Goal 6
Goal 7
Goal 8
-0.5
Goal 9
Goal 10
100 200 300 400 500 600 700 800 900 1000 Goal 11
Generations Goal 12
Goal 13
Goal 14
Goal 15
Goal 16

Goal 2
0.4
Goal 3
Goal 4
Deviations
0.2
Goal 5
0 Goal 6
Goal 7
-0.2 Goal 8
Goal 9
200 400 600 800 1000 Goal 10
Generations Goal 11
Goal 12
Goal 13
Goal 14
Goal 15
Goal 16

0.8
Goal 2
0.6 Goal 3
Goal 4
Deviations
0.4
Goal 5
0.2 Goal 6
Goal 7
0 Goal 8
-0.2 Goal 9
200 400 600 800 1000 Goal 10
Generations Goal 11
Goal 12
Goal 13
Goal 14
Goal 15
Goal 16
FIGURE 4.5 Graphs of Deviations for Linear Model of Level-1, 2 and 3, Kuntal, IJEIT, 2013.
Non-linear Programming graph of level 1 Goal 1

0.4
Goal 2
0.3 Goal 3
0.2 Goal 4
Deviations
Goal 5
0.1 Goal 6
0 Goal 7
-0.1 Goal 8
Goal 9
-0.2 Goal 10
Goal 11
100 200 300 400 500 600 700 800 900 1000 Goal 12
Generations Goal 13
Goal 14
Goal 15
Goal 16
-3
x 10 Non-linear Programming graph of level 2 Goal 1
1.5
Goal 2
1 Goal 3
Goal 4
0.5
Deviations
Goal 5
Goal 6
0
Goal 7
-0.5 Goal 8
Goal 9
-1
Goal 10
-1.5 Goal 11
100 200 300 400 500 600 700 800 900 1000 Goal 12
Generations Goal 13
Goal 14
Goal 15
Goal 16
Non-linear Programming graph of level 3 Goal 1

0.6
Goal 2
0.4 Goal 3
Goal 4
Deviations
0.2 Goal 5
Goal 6
0 Goal 7
Goal 8
-0.2
Goal 9
-0.4 Goal 10
Goal 11
100 200 300 400 500 600 700 800 900 1000 Goal 12
Generations Goal 13
Goal 14
Goal 15
Goal 16
FIGURE 4.6 Deviations for Nonlinear GP Model of Level-1, 2 and 3, Kuntal, IJEIT, 2013
Non-linear Programming graph of level 1
0.4 GA
RST
Deviations
0.2
1 2 3 4 5 6 7 8 9
No of Variables
0.4 GA
RST
Deviations
0.2
1 2 3 4 5 6 7 8 9
No of Variables
0.6 GA
RST
Deviations
0.4
0.2
1 2 3 4 5 6 7 8 9
No of Variables
FIGURE 4.7 Comparison of GA and RST, Kuntal, IJEIT, 2013
4.3.3 Conclusion
Solutions of NLPP (1) and (2) with original and reduced bounds using RST reveals
that the variation in the value of decision variables x1, x2 and x3 are 4.3%, −0.19%
and 20.17% respectively, which leads to 0.68% variation in the objective function
(maximum milk yield). There is an approximately 2% deviation in the objective
function in comparison with results obtained by Pratiksha Sexena using the Kuhn-
Tucker condition.
Solutions of NLPP (1) and (2) using the genetic algorithm (see Tables 4.2 and
4.3) reveals that the variation of decision variables x1, x2 and x3 are 9.3%, 6.9% and
6.08%, respectively, which leads to 27% variation in the objective function, which
is more compared to RST2. Hence, a reduction of bounds gives a better result,
which is comparable with RST2 as there is a deviation of 2% (approximately) in
the objective function with the value obtained by the Kuhn-Tucker method. Using
a wide range of bounds for GA gives better results only after 3000 generations,
whereas after reducing the bounds, the best result is obtained in 500 generations.
After comparing the results obtained by GA with RST2, it reveals that both applied
techniques (RST2 and GA) are nearly the same, with a small difference of 4.6%
(See Figures 4.1, 4.2, 4.3 and 4.4). The possible range of choice variables and goal
characteristics is given in Table 4.4. Therefore, there is a chance of solving the non-
linear model of animal feed formulation using GA with a slight modifcation in the
technique according to our requirements. GA gives adaptability to pick any pos-
sible solutions from the vast domain of the solution set, which adds an extra bonus
for the planner. In this study, the NLP model is solved by using a genetic algo-
rithm for 1000 generations for each level of dairy animals. The result is shown in
Tables 4.6 and 4.7 for the LP and NLP models. Moreover, the comparison of the
controlled random search technique (RST) and the genetic algorithm (GA) for the
nonlinear goal programming problem of each level is shown in Table 4.8. This
investigation centers around a heuristic approach for improvement of the ration
of Indian dairy cows with 10 litres of daily milk yield, considering the nonlinear
weighted sum goal programming formulation. The GA approach is refected as
a better approach in solving nonlinear goal programming problems in compari-
son with the prior methodologies. In the prior work by Shrabani (Shrabani Ghosh
[M.Phil. dissertation] 2011) the linear goal programming problem was solved using
Excel Solver. An endeavour has been made in the present investigation to expand
the linear GPP as a nonlinear weighted sum goal programming problem with pre-
defned priorities at different levels and is solved using the GA approach. Jyothi
Lakshmi likewise made an endeavour to tackle this issue with the controlled ran-
dom search technique (RST). The comparison demonstrates that the GA approach
gives a better outcome as compared to RST (see Figure 4.7). It is likewise seen from
the outcomes appeared in Table 4.6 that Goal-1 is overachieved by 46% to 50% of
variation and Goal-15 is overachieved by 5% to 8.5% of variation for Level-1. In
Level-2, Goal-16 is overachieved by 3.2% to 4.7% of variation. Also, in Level-3
Goal-1 is overachieved by 42.6% to 50.9% of variation. All other goals are achieved
in a thousand generations using GA for the linear model. Figure 4.5 can be alluded
to to see the wholistic perspective of the goal achievements. From Figure 4.6 we
can observe that Goal-1 is overachieved by 16.2% to 20.2% and Goal-15 is over-
achieved by 14.9% to 21.3% of variation for Level-1. For Level-2, all the goals are
achieved, and for Level-3, Goal-1 is overachieved by 46% to 51.8% and Goal-16
is overachieved by 5% to 10% of variation. All other goals are almost achieved in
1000 generations using GA for the NLP model. Thus, by considering these out-
comes for the LP model, our tool gives more profcient rations by tweaking the
nutritive goals and by taking innocuous deviations into consideration by treating
them as underachievement and overachievement. The connected approach of the
nonlinear weighted goal programming problem ends up being a helpful “motor” to
formulate the least cost ration without any nutritive defciency, which is the basic
downside of the LP. The profcient ration can be improved further by adjusting the
target values of the goals, but this tool may be useful for the assessment of conse-
quences due to globalization impacts (input price rise, price volatility and environ-
mental as well as climate change aspects).
e
nc
ie
k sc
oc
st
li ve
e/
t.m
s ://
tp
ht
5
Least Cost Feed Formulation for Dairy
Cattle during Pregnancy by Using
Real Coded Genetic Algorithm
An Application
CONTENTS
5.1 Introduction.................................................................................................. 45
5.1.1 Data Collection and Methodology .................................................. 46
5.1.1.1 LP Model for Cattle-1 ...................................................... 50
5.2 Implementation ............................................................................................ 52
5.2.1 Results and Discussions................................................................... 53
5.2.2 Conclusion ....................................................................................... 59
5.3 A Goal Programming Approach to Ration Formulation Problem for
Indian Dairy Cows ....................................................................................... 60
5.3.1 Goal Programming Approach ......................................................... 62
5.3.1.1 GP Model-1 ...................................................................... 63
5.3.1.2 GP Model-2 ...................................................................... 63
5.3.1.3 GP Model-3 ...................................................................... 64
5.3.2 Real Coded Genetic Algorithm with Hybrid Function ................... 65
5.3.3 Results ............................................................................................. 66
5.3.4 Discussion........................................................................................ 67
5.3.5 Conclusion ....................................................................................... 68
5.1 Introduction
In the Mandya district of Karnataka, dairy farmers have limited feedstuffs.
Therefore, to fulfll nutrient requirements in terms of crude protein (CP), total
digestible nutrients (TDN), calcium (Ca) and phosphorus (P), it was necessary
to adopt a software program. There are many techniques/software that were
adopted to fulfll the nutrient requirements of dairy animals (www.scialert.net);
however, there is no specifc technique suitable for formulating least cost rations
for dairy animals. Hence, a comparative study was planned to fnd a suitable tech-
nique which is farmer friendly and can be used at the dairy farm level. The focus
of this chapter is to provide a least cost diet formulation based on the nutrient
DOI: 10.1201/9781003231714-5 45
requirements of cattle which has been calculated according to the Indian Council
of Agricultural Research (ICAR-NIANP, 2013) and National Research Council
(NRC, 2001) standards during the third trimester of pregnancy. A linear program-
ming model (LP) was formulated for three different types of cattle with body
weight of 500 kg each and the requirement of a 10-litre milk yield with 4% of
fat content, based on the seventh, eighth and ninth months of pregnancy on a dry
matter basis. The LP model has been solved by a general reduced gradient (GRG)
nonlinear, an evolutionary algorithm (EA), a simplex LP method and a real coded
genetic algorithm (RGA) with and without seeding the random number genera-
tions. As far as the proft is concerned for the farmers, the cost of the feed plays
an important role.
In 1951, Waugh applied linear programming technique developed by
Koopmans (Waugh, 1951) to provide the solution of least cost dairy feed.
Therefore, for the past seven decades conventional linear programming
method was very popular to solve animal diet problems (Rehman, 1984;
Oladokun and Johnson, 2012). In 2001, Tozer applied a multiple-objective
technique to minimize the nutrient excretion from dairy cows through incor-
poration of a nutrient excretion function into ration formulation. In 2002,
Zhang also applied multiple-objective programming to the feed formulation
of a broiler grower ration with the objective of minimizing the nutrient vari-
ance and minimizing the ration cost. Many researchers have introduced the
use of computer software and Excel Solver for solving linear programming
problems (Folayan et al., 2008; Zgajnar et al., 2009; Sebastian et al., 2008;
Radhika and Rao, 2010). Solving diet problems using the LP model math-
ematically only deals with primary technical issues, which can have a weak
relationship with the commercial complication of reducing or increasing
the change between costs and feeding over time. To overcome this problem
various types of methodologies such as goal programming, multi-objective
programming, nonlinear programming etc. have been used for many years
(Rehman and Romero, 1984, 1987; Hertzler, 1987; Afrouziyeh et al., 2010;
Saxena, 2011a). It has been shown that the linear programming approach indi-
cates how best to combine available local ingredients effectively to formulate
the least cost ration for broilers (Al-Deseit, 2009). Ghosh in 2014 combined
LP with goal programming (GP) for feed formulation, displaying the dip in
cost as a major advantage of using the goal-programming approach (Ghosh
et al., 2014).
5.1.1 Data Collection and Methodology

The present work is based on primary data collected from the Mandya district of
Karnataka as per ICAR-NIANP 2013 and NRC 2001 standards (ICAR-NIANP,
2013). Briefy, the cows during the third trimester of pregnancy need a balanced
ration for body maintenance, milk production (with a minimum of 4% of fat
content) and fetus growth. Therefore, to formulate the low-cost diet for these
cows is a major challenge for farmers. So we have introduced LP models at
Feed Formulation Using RGA 47
three different months of pregnancy for different body weights. As the growth
of the fetus increases signifcantly after six months of pregnancy, the nutrient
requirements of cattle at the seventh, eighth and ninth month of pregnancy are
presented in Tables 5.1 and 5.2. The estimation of nutrient requirements dur-
ing late pregnancy requires accurate values for rates of nutrient accretion in
conceptus tissues (ICAR-NIANP, 2013). The ration was formulated using the
following steps:
Step-1 (calculation of nutrient requirement): The nutrient requirements

of Cattle-, 2, and 3 are calculated based on body weight, milk yield, fat
percentage and pregnancy status. Dry matter intake, crude protein and
total digestible nutrients are calculated according to ICAR-NIANP 2013,
phosphorus and calcium according to NRC 2001 standards on a dry matter
basis (Table 5.1). Different categories of cattle are considered as follows:
TABLE 5.1
Nutrient Requirements of Cattle-1, 2 and 3
Body Milk yield Pregnancy
Level weight (kg) (in litres) Fat % (in months)
Cattle 1 500 10 4 7
Cattle 2 500 10 4 8
Cattle 3 500 10 4 9
Step-2 (selection of ingredients): Based on primary data collected in the

Mandya district of Karnataka, locally available roughages, concentrate and
minerals are listed in Table 5.2. Among these ingredients, commonly avail-
able dry roughages are paddy straw and Ragi straw. Commonly available
green roughages are Bajra X Napier grass (Co-4), multi-cut sorghum (Co-FS
29) and maize fodder. The nutrient content of the feed stuffs is shown in
Table 5.2, and nutrient contents of feedstuffs are shown in Table 5.3.
TABLE 5.2
Selected Feedstuffs
Roughages Concentrates Minerals
Paddy straw Maize Calcite

Bajra X Napier Co-4 grass Soya de-oiled cake (DOC) Mineral mixture (MM)
Maize fodder Copra DOC Dicalcium phosphate (DCP)
Co-FS 29 sorghum Cotton DOC Salt
Ragi straw Wheat bran
Gram chunnies
Cotton seed
Concentrate mix Type-1
TABLE 5.3
Nutrient Content of Feedstuffs on DM basis
Feeds DM (%) CP (kg) TDN (kg) Ca (kg) P (kg) Cost (Rs)
Roughage
Paddy straw 90 0.0513 0.4 0.0018 0.0008 5
Co-4 grass 25 0.08 0.52 0.00144 0.0009 3
Maize fodder 25 0.08 0.6 0.0053 0.0014 3
Co-FS 29 sorghum 90 0.07 0.5 0.0012 0.0009 3
Ragi straw 90 0.06 0.42 0.0015 0.0009 3
Concentrates
Maize 90 0.081 0.792 0.00018 0.0027 17
Soya DOC 90 0.42 0.7 0.00018 0.00225 38
Copra DOC 90 0.22 0.7 0.00036 0.009 23
Cotton DOC 90 0.32 0.7 0.00036 0.009 23
Wheat bran 90 0.12 0.7 0.01067 0.00093 17
Gram chunnies 90 0.17 0.6 0.000108 0.00234 12
Cotton seed 90 0.16 1.1 0.003 0.0062 21
Conc. mix Type-I 90 0.22 0.7 0.005 0.0045 17
Calcite 97 0 0 0.36 0 4
MM 90 0 0 0.32 0.06 50
DCP 90 0 0 0.24 0.16 28
Salt 90 0 0 0 0 5
Step-3 (fxing the constraints): To obtain the least cost feed we have some con-
straints for each nutrient, and it is unique for each animal. The minimum and
maximum level of crude protein (CP 9.82–11%), total digestible nutrient (TDN
46.73–51%), calcium (0.38–0.8%), phosphorus (0.23–0.5%), roughage (40–
80%) and concentrates (20–70%) was calculated on total dry matter intake for
each type of cattle. The calculated constraints for each animal are as follows:
Constraints for Cattle 1:
°
17
1. i=1
DMI i = 16.75 kg
2. 1.644 kg ˛ °13
i=1 CPi ˛1.8425 kg
3. 7.83kg ˛ °13
i=1 TDN i ˛ 8.5425 kg
4. 0.065 kg ˛ °17
i=1 Cai ˛ 0.134 kg
5. 0.04 kg ˛ °17
i=1 Pi ˛ 0.08375 kg
5
6. 6.7 kg ˛ ° i=1 Roughagesi ˛13.4 kg
7. 3.35 kg ˛ °17
i=6 Concentrates i ˛11.725 kg

1. °17
i=1 DMI i = 16.89 kg
2. 1.691 kg ˛ °13
i=1 CPi ˛1.8579 kg
3. 7.793 kg ˛ °13
i=1 TDN i ˛ 8.6139 kg
4. 0.065 kg ˛ °17
i=1 Ca ˛ 0.13512 kg
5. 0.04 kg ˛ °17
i=1 Pi ˛ 0.08445 kg
5
7. 3.378 kg ˛ °17
1. °17
i=1 DMI i = 17.03 kg
2. 1.738 kg ˛ °13
i=1 CPi ˛1.8733 kg
3. 8.03kg ˛ °13
i=1 TDN i ˛ 8.6853 kg
4. 0.065 kg ˛ °17
i=1 Cai ˛ 0.13624 kg
5. 0.04 kg ˛ °17
i=1 Pi ˛ 0.08515 kg
5
7. 3.406 kg ˛ °17
Step-4 (fnding results): In this step, we fnd the least cost of “feedstuffs”
after fulflling constraints assigned by solving the linear programming
model with various mathematical tools and then translating the obtained
results as a recommendation to the farmers. The schematic diagram of the
methodology followed is shown in Figure 5.1.
Data Entry
Animal Weight Production
Set Requirement
Maintenance Production
on DM Basis
Set Feed Resources
Least cost feed formulation
LP GRG Evolutionary Real Code Genetic Algorithm

Simplex Non-Linear Algorithm (EA)
RGA (1) RGA (17)
Converted to
Result out put on DM Basis
Fresh Basis
FIGURE 5.1 Schematic Diagram Representing Methodology Followed for Least Cost Feed
Formulation of Dairy Cattle; Kuntal, AJAVA, 2016.
5.1.1.1 LP Model for Cattle-1

Minimize Z = 5x1 + 3x2 + 3x3 + 3x4 + 3x5 +17x6 + 38x7 + 23x8
+ 23x9 +17x10 +12xx11 + 21x12 +17x13 + 4x14
+ 50x15 + 28x16 + 5x17
Subjected to:
1. x1 + x2 + x3 + x4 + x5 + x6 + x7 + x8 + x9 + x10 + x11 + x12 + x13 + x14
+ x15 + x16 + x17
1 = 16.75
2. 1.644 ˛ 0.0513 x1 + 0.08 x2 + 0.08x3 + 0.07 x4 + 0.06x5 + 0.081x6 + 0.42x7

+ 0.22 x8 + 0.32 x9 + 0.12 x10 + 0.17 x11 + 0.16 x12 + 0.22 x13 ˛1.8425
3. 7.83 ˛ 0.4x1 + 0.52 x2 + 0.6 x3 + 0.5x4 + 0.42 x5 + 0.792x6 + 0.7 x7 + 0.7 x8
+ 0.7x
. 9 + 0.7 x10 + 0.6 x11 +1.1x12 + 0.7 x13 ˛ 8.5425
4. 0.065 ˛ 0.0018x1 + 0.00144 x2 + 0.0053x3 + 0.0012 x4 + 0.0015x5
+ 0.00018x
8 6 + 0.00018 x7 + 0.00036 x8 + 0.00036 x9 + 0.01067 x10
+ 0.000108x11 + 0.003 x12 + 0.005x13 + 0.36 x14 + 0.32 x15
+ 0.24x16 ˛ 0.134
5. 0.04 ˛ 0.0008 x1 + 0.0009 x2 + 0.0014 x3 + 0.0009 x4 + 0.0009 x5 + 0.0027x6
+ 0.00225x7 + 0.009x8 + 0.009x9 + 0.00093x10 + 0.00234x11 + 0.0062 x122
+ 0.0045x13 + 0.06x15 + 0.16 x16 ˛ 0.08375
6. 6.7 ≤ x1 + x2 + x3 + x4 + x5 ≤ 13.4
7. 3.35 ˛ x6 + x7 + x8 + x9 + x10 + x11 + x12 + x13 + x14 + x15 + x16
+ x17 ˛11.725
Bounds on variables: 0.8375 ˛ x1 ˛ 4.1875, 0.8375 ˛ x2 ˛ 4.1875, 0.8375 ˛ x3

˛ 4.1875, 0.8375
. ˛ x4 ˛ 4.1875, 0.8375 ˛ x5 ˛ 4.1875, 0.8375 ˛ x6 ˛ 3.35, 0 ˛ x7
˛ 4.1875, 0 ˛ x8 ˛ 4.1875, 0 ˛ x9 ˛ 3.35, 0.8375 ˛ x10 ˛1.675, 0 ˛ x11 ˛ 3.35,
0 ˛ x12 ˛ 0.8375, 0.8375 ˛ x13 ˛ 3.35, 0 ˛ x14 ˛ 0.1675, 0.067 ˛ x15 ˛ 0.08375,
0 ˛ x16 ˛ 0.0335, 0.134 ˛ x17 ˛ 0.1675

Minimize Z = 5x1 + 3x2 + 3x3 + 3x4 + 3x5 +17x6 + 38x7 + 23x8 + 23x9
2 11 + 21x12 + 17x13 + 4x14 + 50x15 + 28x16 + 5x17
+17x10 +12x
Subjected to:
1. x1 + x2 + x3 + x4 + x5 + x6 + x7 + x8 + x9 + x10 + x11 + x12 + x13 + x14 + x15
+ x16 + x17 = 16.89
2. 1.691˛ 0.0513x1 + 0.08 x2 + 0.08x3 + 0.07 x4 + 0.06x5 + 0.081x6 + 0.42 x7
+ 0.22x8 + 0.32x9 + 0.12x10 + 0.17 x11 + 0.16x12 + 0.22x13 ˛1.8579
3. 7.93 ˛ 0.4x1 + 0.52 x2 + 0.6 x3 + 0.5x4 + 0.42 x5 + 0.792x6 + 0.7 x7 + 0.7 x8
+ 0.7x
. 9 + 0.7 x10 + 0.6 x11 +1.1x12 + 0.7 x13 ˛ 8.6139
4. 0.065 ˛ 0.0018x1 + 0.00144 x2 + 0.0053x3 + 0.0012 x4 + 0.0015x5
+ 0.00018x
8 6 + 0.00018 x7 + 0.00036 x8 + 0.00036 x9 + 0.01067 x10
+ 0.000108x11 + 0.003 x12 + 0.005x13 + 0.36 x14 + 0.32x15 + 0.24x16
˛ 0.13512
5. 0.04 ˛ 0.0008x1 + 0.0009x2 + 0.0014 x3 + 0.0009x4 + 0.0009 x5 + 0.0027x6
+ 0.00225x7 + 0.009x8 + 0.009x9 + 0.00093x10 + 0.00234x11 + 0.0062 x122
+ 0.0045x13 + 0.06x15 + 0.16 x16 ˛ 0.08445
6. 6.756 ˛ x1 + x2 + x3 + x4 + x5 ˛ 13.512
7. 3.378 ˛ x6 + x7 + x8 + x9 + x10 + x11 + x12 + x13 + x14 + x15 + x16
+ x17 ˛11.823

˛ 4.2225, 0.84445 ˛ x4 ˛ 4.2225, 0.8445 ˛ x5 ˛ 4.2225, 0.8445 ˛ x6 ˛ 3.378,
0 ˛ x7 ˛ 4.22225, 0 ˛ x8 ˛ 4.2225, 0 ˛ x9 ˛ 3.378, 0.8445 ˛ x10 ˛ 1.689, 0 ˛ x11
˛ 3.3778, 0 ˛ x12 ˛ 0.8445, 0.8445 ˛ x13 ˛ 3.378, 0 ˛ x14 ˛ 0.1689, 0.06756 ˛ x15
˛ 0.8445, 0 ˛ x16 ˛ 0.03378, 0.13512 ˛ x17 ˛ 0.1689

Minimize Z = 5x1 + 3x2 + 3x3 + 3x4 + 3x5 +17x6 + 38x7 + 23x8 + 23x9
2 11 + 21x12 + 17x13 + 4x14 + 50x15 + 28x16 + 5x17
+ 17x10 +12x
Subjected to:
1. x1 + x2 + x3 + x4 + x5 + x6 + x7 + x8 + x9 + x10 + x11 + x12 + x13 + x14 + x15
+ x16 + x17 = 17.03
2. 1.738 ˛ 0.0513 x1 + 0.08 x2 + 0.08x3 + 0.07 x4 + 0.06x5 + 0.081x6 + 0.42x7
+ 0.22 x8 + 0.32 x9 + 0.12 x10 + 0.17 x11 + 0.16 x12 + 0.22 x13 ˛1.8733
3. 8.03 ˛ 0.4x1 + 0.52 x2 + 0.6 x3 + 0.5x4 + 0.42 x5 + 0.792x6 + 0.7 x7 + 0.7 x8

+ 0.7x
. 9 + 0.7 x10 + 0.6 x11 +1.1x12 + 0.7 x13 ˛ 8.6853
4. 0.065 ˛ 0.0018x1 + 0.00144 x2 + 0.0053x3 + 0.0012 x4 + 0.0015x5
+ 0.00018x
8 6 + 0.00018 x7 + 0.00036 x8 + 0.00036 x9 + 0.01067 x10
+ 0.000108x11 + 0.003 x12 + 0.005x13 + 0.36 x14 + 0.32x15 + 0.24x16
˛ 0.13624
5. 0.04 ˛ 0.0008x1 + 0.0009x2 + 0.0014 x3 + 0.0009x4 + 0.0009 x5 + 0.0027x6
+ 0.00225x7 + 0.009x8 + 0.009x9 + 0.00093x10 + 0.00234x11 + 0.0062 x122
+ 0.0045x13 + 0.06x15 + 0.16 x16 ˛ 0.08515
6. 6.812 ˛ x1 + x2 + x3 + x4 + x5 ˛ 13.624
7. 3.406 ˛ x6 + x7 + x8 + x9 + x10 + x11 + x12 + x13 + x14 + x15 + x16
+ x17 ˛11.921

˛ 4.2225, 0.8515
1 ˛ x4 ˛ 4.2225, 0.8515 ˛ x5 ˛ 4.2225, 0.8515 ˛ x6 ˛ 3.406, 0 ˛ x7
˛ 4.25
575, 0 ˛ x8 ˛ 4.2575, 0 ˛ x9 ˛ 3.406, 0.8515 ˛ x10 ˛ 1.703, 0 ˛ x11 ˛ 3.4006,
0 ˛ x12 ˛ 0.8515, 0.8515 ˛ x13 ˛ 3.406, 0 ˛ x14 ˛ 0.1703, 0.06812 ˛ x15 ˛ 0.08515,
0 ˛ x16 ˛ 0.03406, 0.13624 ˛ x17 ˛ 0.1703
5.2 Implementation
As suggested by Ghosh et al., (2014), we have explored the possibility of using
various techniques including genetic algorithm (GA) to solve the feed formulation
problem (refer to Chapter 4). The success of GA is in its evolution process. More
recently, real coded genetic algorithm (RGA) has been used in numerical optimiza-
tion (Herrera et al., 1995; Ono et al., 2003), and most of the optimization is done in
real space (continuous space). However, constraints are rigid; the use of RGA over-
comes the low resolution of solution weakness of binary coded GA (Baron et al.,
2002; Achiche et al., 2002). The outline of RGA is as follows:
Step-1: Genetic algorithm starts with frst creating the initial population
randomly. In MATLAB 2013, if we set the population size, then the ini-
tial population generated by the algorithm is same as population size by
default, so to solve our linear programming problem we have generated a
500-population size.
Step-2: Sequence of new population and next generation: This creates
a sequence of new population using RGA operators, namely crossover,
mutation and selection. At every step of RGA, the current population is
used to form new offspring that create the next generation. The current
population selected is called parents and the generated new population

is called offspring. RGA selects the best ft population as parents to
generate new offspring, which is done by selection operator. RGA pro-
vides three types of offspring for the next generation, namely crossover
offspring, mutation offspring and elite offspring. Crossover offspring
are created by the crossover operator, which shows how GA combines
two parents to undergo crossover for creating crossover children. In our
case, we have used a heuristic crossover method to generate the chil-
dren, because it moves from worst parent to past best parent. We have
used the default ratio 1.2, where this ratio indicates how far offspring is
from the better parent. If P1 and P 2 are two parents, where P1 is having
a better ftness value, then child = P 2 + Ratio (P1 − P 2). Mutation off-
spring can be created by randomly changing individuals in the popula-
tion. It is an important operator to maintain diversity and search broader
spaces. In our study we have used adaptive feasible mutation because it
performs better on linear constraints. It randomly generates directions
that are adaptive with respect to its previous generation. This method
chooses the direction and step length that satisfes linear constraints.
Elite offspring are created based on positive integer values after cross-
over or mutation operators, which specify how many best ft offspring
are guaranteed to survive in each generation. We have kept the elite
count at 2, and it is a good operator for the convergence point of view;
keeping a high value of elite count does not guarantee good conver-
gence, as we need some worst ft solution every generation as well. We
used a tournament selection procedure, which selects parent by select-
ing tournament size at random and then selecting best ft parents. We
kept default tournament size at 4. We also used the initial penalty as
10 and penalty factor as 100, where initial penalty specifes the initial
penalty used for nonlinear constraint and the penalty will increase the
penalty parameter whenever there are constraint violations or whenever
the problem is not providing accuracy. Generally, this option of RGA
is used for nonlinear constraints, but as we have rigid constraints and
bounds, we have used penalty factors as an option to optimize the prob-
lem (Kuntal et al., 2016, 2018).
Step-3: Stopping criteria of the algorithm: We have used generation as a
stopping criterion, where once the maximum generation is reached, the
algorithm stops and the exit fag gives the reason why the algorithm has
stopped.
5.2.1 Results and Discussions

The results obtained by various techniques, viz. LP simplex, GRG nonlinear,
evolutionary algorithm (EA) and real coded genetic algorithm (RGA), with seed-
ing the random numbers for least cost ration by LP model are presented in Tables
5.4, 5.6 and 5.8 for Cattle-1, 2 and 3 on a DM basis, respectively. LP model-1, 2
TABLE 5.4
Least Cost of Ration by Various Techniques for Cattle-1 on DM Basis
Simplex & EA with RGA with RGA with
Variables GRG nonlinear seeding RNG (1) RNG (17)
x1 3.3184369 3.46200246 3.1889 2.8845

x2 3.1048270 1.88904534 3.0043 2.8115
x3 0.8375 1.13120299 0.8375 0.8375
x4 4.1875 0.87935706 1.5299 1.7039
x5 0.8375 1.87331391 3.7707 4.0965
x6 0.8375 1.15084533 0.8375 0.8375
x7 0 0.14970428 0.2189 0.2203
x8 0 0.67154322 0 0
x9 1.5497360 0.38100715 1.2351 1.2311
x10 0.8375 1.43585220 0.8375 0.8375
x11 0 0.85084407 0 0
x12 0 0.14824419 0 0
x13 0.8375 2.11359082 0.8375 0.8375
x14 0.1675 0.15013463 0.1675 0.1675
x15 0.067 0.06981443 0.0837 0.0838
x16 0 0.01619651 0.0335 0.0335
x17 0.1675 0.16741651 0.1675 0.1675
CP 1.644 1.82458530 1.6440 1.644
TDN 8.5425000 9.18352968 8.5425 8.5425
Calcium 0.1176172 0.12612679 0.1309 0.1308
Phosphorus 0.0419254 0.04198480 0.0400 0.04
Roughage 12.285764 9.23492177 12.3313 12.3339
Concentrate 4.4642360 7.30519339 4.4187 4.4162
DMI 16.75 16.5401151 16.7500 16.7501
Least cost 126.708094 163.136196 129.4403 128.8009
TABLE 5.5
Least Cost of Ration by Various Techniques for Cattle-1 on As Fresh Basis (LP Model)
Simplex & RGA without RGA with Real coded GA
Variables GRG nonlinear seeding RNG RNG (1) with RNG (17)
x1 3.687152 3.084888889 3.543222222 3.205

x2 12.41931 10.4672 12.0172 11.246
x3 3.35 3.35 3.35 3.35
x4 4.652778 2.126555556 1.699888889 1.893222222
x5 0.930556 4.652777778 4.189666667 4.551666667
x6 0.930556 0.930555556 0.930555556 0.930555556
x7 0 0.249 0.243222222 0.244777778
x8 0 0 0 0
x9 1.721929 1.365666667 1.372333333 1.367888889
Simplex & RGA without RGA with Real coded GA

Variables GRG nonlinear seeding RNG RNG (1) with RNG (17)
x10 0.930556 0.930555556 0.930555556 0.930555556

x11 0 0 0 0
x12 0 0 0 0
x13 0.930556 0.930555556 0.930555556 0.930555556
x14 0.17268 0.172680412 0.172680412 0.172680412
x15 0.074444 0.093111111 0.093 0.093111111
x16 0 0.037222222 0.037222222 0.037222222
x17 0.186111 0.186111111 0.186111111 0.186111111
TMR as fresh basis 29.98663 28.57688041 29.69621375 29.13934708
(in kg)
Least cost in Rs on 174.8999 172.8637661 177.0643216 174.688055
as fresh basis
Least cost (Rs/kg) 5.832596 6.049078 5.962522 5.99492
TABLE 5.6
Least Cost of Ration by Various Techniques for Cattle-2 on DM Basis (LP Model)
Simplex & EA with RGA With RGA With
x1 3.5268827 3.4909386 3.4237 4.2225

x2 2.7694622 1.9048344 2.3255 0.8445
x3 0.8445 1.1406578 0.8445 0.8445
x4 0.8445 0.8867069 1.4873 3.8991
x5 4.2225 1.8889715 3.8958 2.4448
x6 0.8444 1.1604643 0.8445 0.8445
x7 0 0.1509555 0.3213 0.4701
x8 0 0.6771561 0.0563 0.0933
x9 1.7280454 0.3841917 1.0548 1.0817
x10 0.8444 1.4478534 0.8445 0.8445
x11 0 0.8579556 0.4912 0
x12 0 0.1494833 0.0000 0
x13 0.8445 2.1312567 0.8445 0.8445
x14 0.1689 0.1513895 0.1689 0.1689
x15 0.06756 0.070398 0.0844 0.0844
x16 0.0154497 0.0163319 0.0338 0.0338
x17 0.1689 0.1688158 0.1689 0.1689
CP 1.691 1.8398356 1.6910 1.691
TDN 8.6139 9.2602875 8.6139 8.6139
Calcium 0.1222969 0.127181 0.1319 0.1315
Phosphorus 0.04 0.0423357 0.0401 0.04
Roughage 12.207845 9.3121092 11.9768 12.2554
Concentrate 4.6821551 7.3662517 4.9131 4.6346
DMI 16.89 16.678361 16.8899 16.89
Least cost 131.81914 164.49972 136.1944 139.856
TABLE 5.7
Least Cost of Ration by Various Techniques for Cattle-2 on As Fresh Basis (LP Model)
Simplex & RGA without RGA with RGA with
Variables GRG nonlinear seeding RNG RNG (1) RNG (17)
x1 3.919248 3.284889 3.804111 4.691667

x2 11.07516 10.1896 9.302 3.378
x3 3.378 3.378 3.378 3.378
x4 0.938333 1.974222 1.652556 4.332333
x5 4.691667 4.691667 4.328667 2.716444
x6 0.938333 0.938333 0.938333 0.938333
x7 0 0.420222 0.357 0.522333
x8 0 0 0.062556 0.103667
x9 1.920089 1.305222 1.172 1.201889
x10 0.938333 0.938333 0.938333 0.938333
x11 0 0 0.545778 0
x12 0 0 0 0
x13 0.938333 0.938333 0.938333 0.938333
x14 0.174124 0.174124 0.174124 0.174124
x15 0.075067 0.093889 0.093778 0.093778
x16 0.017163 0.037556 0.037556 0.037556
x17 0.187667 0.187667 0.187667 0.187667
TMR on as fresh 29.19152 28.55206 27.91079 23.63246
basis in kg)
Least cost in Rs on 177.7315 178.3493 178.7446 169.9794
as fresh basis
Least cost (Rs/kg) 6.088463 6.24646 6.40414 7.192624
TABLE 5.8
Least Cost of Ration by Various Techniques for Cattle-3 on DM Basis (LP Model)
x1 3.85658052 3.5198747 3.1263 3.5056
x2 2.33299069 1.9206234 2.323 2.4336
x3 0.8514 1.1501126 0.8515 0.8515
x4 0.8515 0.8940567 1.82 1.7554
x5 4.2575 1.9046290 4.2393 3.8079
x6 0.8514 1.1700833 0.8515 0.8515
x7 0 0.1522068 0.5362 0.5378
x8 0 0.6827690 0 0
x9 1.91220552 0.3873762 1.1195 1.1239
x10 0.8514 1.4598545 0.8515 0.8515
x11 0 0.8650671 0 0
x12 0 0.1507223 0 0

x13 0.8515 2.1489224 0.8515 0.8515

x14 0.17025892 0.1526443 0.1703 0.1703
x15 0.06812 0.0709814 0.0851 0.0852
x16 0.00503889 0.0164672 0.0341 0.0341
x17 0.1703 0.1702151 0.1703 0.1703
CP 1.73800000 1.8550858 1.7380 1.738
TDN 8.68530003 9.3370454 8.6854 8.6853
Calcium 0.12070605 0.1282351 0.1327 0.1329
Phosphorus 0.04 0.0426866 0.0400 0.04
Roughage 12.1499712 9.3892965 12.3601 12.354
Concentrate 4.88012333 7.4273100 4.6700 4.676
DMI 17.0300945 16.816606 17.0301 17.03
Least cost 136.644426 165.86324 139.6257 140.5297
TABLE 5.9
Least Cost of Ration by Various Techniques for Cattle-3 on As Fresh Basis
Simplex & RGA without RGA with RGA with
Variables GRG nonlinear seeding RNG RNG (1) RNG (17)
x1 4.285569 3.44955555 3.4736666 3.895111

x2 9.327939 8.3764 9.292 9.7344
x3 3.406 3.406 3.406 3.406
x4 0.946111 2.31666666 2.0222222 1.9504444
x5 4.730556 4.68933333 4.7103333 4.231
x6 0.946111 0.94611111 0.9461111 0.946111111
x7 0 0.60211111 0.5957777 0.597555556
x8 0 0 0 0
x9 2.124722 1.24244444 1.2438888 1.248777778
x10 0.946111 0.94611111 0.9461111 0.946111111
x11 0 0 0 0
x12 0 0 0 0
x13 0.946111 0.94611111 0.9461111 0.946111111
x14 0.175567 0.17556701 0.1755670 0.17556701
x15 0.075689 0.09466666 0.0945555 0.094666667
x16 0.005592 0.03788888 0.0378888 0.037888889
x17 0.189222 0.18922222 0.1892222 0.189222222
TMR as fresh 28.1053 27.4181892 28.079455 28.39896701
basis (in kg)
Least cost in Rs 179.3693 180.763690 182.59771 184.5643569
on as fresh basis
Least cost 6.382045 6.592838 6.502894 6.498981
(Rs/kg)
TABLE 5.10
Net Proft Per Cattle in INR to Farmers for 10 Litres of Milk on As Fresh Basis Each
Day
Milk rate Net proft by Net proft by Net proft by Net proft by
in dairy simplex LP & RGA without RGA with RGA with
Level (INR) GRG nonlinear seeding RNG (1) RNG (17)
Cattle-1 22.6 51.1001 53.13623 48.93568 51.311945

Cattle-2 22.6 48.2685 47.6507 47.2554 56.0206
Cattle-3 22.6 46.6307 45.23631 43.40229 41.4356431
Mean ± S. E 48.67±1.31 48.67±2.34 46.53±1.64 49.59±4.30
P value 0.868 NS
NS–Non-signifcance: P > 0.05. No signifcance difference exists between methods.
and 3 for each type of cattle have 17 variables, which are too complex for fnding
an optimal solution graphically; hence, we used the LP simplex method, which
provides an iterative algorithm to locate the corner points systematically until
we get an optimal solution. A GRG nonlinear algorithm with forward differenc-
ing deals with problems involving decision variables and nonlinear constraints.
Forward differencing uses a single point that is slightly different from current
value to fnd the derivative; hence, while solving LP models in Excel Solver, it
will not compute derivatives again and again, and it continues to estimate the
solution along the straight line instead of recalculating the changing gradients.
When GRG fnds a solution, this means that solver has found a valley for mini-
mizing the objective function after satisfying the Karush-Kuhn-Tucker (KKT)
conditions for local optimality, and there is no other possible solution for deci-
sion variables (feedstuffs) near to current values. The evolutionary algorithm
(EA) technique is also used to solve LP model by seeding the random number
generator. In the present study, the EA method with seeding technique could not
fnd a lower least cost than the other three methods because costlier nutrients
like the crude protein level in fnal feed was higher (1.82 kg) in EA than other
methods (1.644 kg). Similarly, the TDN content is higher in the EA method than
other methods. These could be the reason for higher costs in EA (INR 163.14)
than the other methods (GRG-126.71; LP- 126.71: RGA seeding 1–129.44 and
RGA seeding 17–128.8). It is evident that the EA method had a not so accurate
least cost feed formulation method because it uses only mutation as a parameter
to improve the diversity of the population in every generation. The EA method is
heuristic in nature and uses only mutation as a parameter to improve the diver-
sity of the problem; hence, while solving the problem the constraints are not
satisfed properly and almost all the constraints are violated. If we do not seed
the random number, then we get the wide ranges of solutions in every run most
of the time, but unfortunately in this study we have not achieved the optimal
solution using EA in MS Excel Solver due to the rigidity of the constraints. The
RGA with seeding technique is applied to overcome the limitation of EA. RGA
is also a heuristic technique which works on the principle of survival of best ft
and tournament selection. Adaptive feasibility mutation and heuristic crossover

is used along with elitism to solve the LP model for Cattle-1, 2 and 3. Though
the seeding technique gives a near optimal answer, we prefer to provide a wide
ranges of solutions to farmers in which the values of feedstuffs changes in every
run where all the constraints will satisfy. As a farmer needs the total mixed ration
to feed the cattle on an “as fresh basis”, we have converted the least cost and
total mixed ration obtained by all techniques to as fresh basis, and the results
are given in Tables 5.5, 5.7 and 5.9. As per the ftment in Table 5.4, each animal
of the Cattle-1 category requires about 16.75 kg of dry matter from various feed
ingredients which should contain 1644 g of protein, 8542 g of TDN, 117.6 g of
calcium and 42 g of phosphorus, and these nutrients can be met from 12.29 kg
of roughages and 4.464 kg of concentrate on dry matter. The corresponding total
mixed ration (TMR) cost on dry matter basis is Rs 126.70, that is, Rs 7.56 per kg,
because fresh basis is a feed nutrient content with moisture included. After con-
verting to as fresh basis, the feedstuff requirement for Cattle-1 is approximately
30 kg/day, amounting to Rs 5.83/kg using the GRG nonlinear and simplex LP
technique. When RGA was used to try to achieve the expected nutrient require-
ment without seeding and with seeding (RGA1 and RGA17), the least cost ration
obtained was Rs 6.05/kg, Rs 5.96/kg and Rs 5.99/kg, respectively. The detailed
analysis of the ration showed (in Tables 5.3 and 5.4) that it exactly met the
requirement of dry matter, TDN, CP, calcium and phosphorus, and roughage: the
concentrate was also well within the permissible range. Similar analysis has been
done for Cattle-2 (see Tables 5.6 and 5.7), where the least cost ration obtained by
various techniques for as fresh basis is Rs 6.08, Rs 6.25, Rs 6.40 and Rs 7.20 per
kg by LP simplex, GRG nonlinear, RGA1 and RGA17, respectively. For Cattle-3
(see Tables 5.8 and 5.9), the TMR cost for as fresh basis turns out to be Rs 6.38,
Rs 6.59, Rs 6.50 and Rs 6.50 per kg, respectively. Table 5.9 shows the net proft
that farmers can make per cattle in Indian rupees for 10 litres of milk each day
on an as fresh basis by all techniques. A one-way ANOVA test at 5% level of
signifcance has been performed for the “Null hypothesis: there is no signifcant
difference between techniques”, the test reveals that since the P value is greater
than 0.05, there is no signifcance difference between the techniques. Hence, it is
proved that the real parameters-based genetic algorithm (www.scialert.net) can
be effectively used for low-cost diet formulation of dairy cattle.
5.2.2 Conclusion
This study addresses the use of RGA as a tool to provide good quality feed mix
to dairy cattle for better health and milk production. All the techniques – viz.
LP simplex, GRG nonlinear, evolutionary algorithm (EA) and real coded genetic
algorithm (RGA) with seeding the random numbers – for least cost ration are per-
forming equally. Hence, it is concluded that RGA (www.scialert.net) will give low-
cost diet formulation for dairy cattle. However, the fxing of constrains and use of
code for making software is also be considered while choosing the techniques for
making least cost feed formulation. Further detailed research with various spe-
cies of animals and with different physiological needs may require to fne-tune
the techniques for farmer use. We also tried to cut down cost of TMR without any
shortage of nutrients in the diet.
5.3 A Goal Programming Approach to Ration

Formulation Problem for Indian Dairy Cows
Due to the limitation of feedstuffs in the Mandya district of Karnataka, small
dairy farmers faced many problems to feed a balanced, least cost diet to dairy cat-
tle. From earlier research it was clear that the productivity of cattle maintained by
different dairy farmers was lower, and this is mainly due to limited resources for
feeding and small farmers not having proper knowledge or resources to provide a
low-cost balanced ration to cattle. Therefore, there is a need to focus on minimiz-
ing the diet cost by upgrading scientifc dairy farming practices. Several tech-
niques are in use for animal diet formulation, but a successful application of a soft
computing technique to improve the quality of the solution is always preferred, as
the rigidity of the functions in a linear programming problem (LPP) can be easily
handled. Hence, to meet the nutrient requirement, a primitive goal programming
(GP) model for three categories of dairy cattle weighing 500 kg each and yielding
10 litres of milk with 4% fat content during the seventh, eighth and ninth month
of pregnancy has been formulated by dividing the goals into a set of priorities.
In our earlier work (Kuntal et al., 2016), LP models for three categories of dairy
cattle has been formulated and solved by the simplex method, GRG nonlinear
method, EA method and real coded genetic method (RGA). In the present work,
a GP model has been developed by dividing each goal into a set of priorities for
all three categories of animal, as there are two high priority objectives – least cost
and dry matter intake – to be achieved if possible. This GP model is solved by a
real coded genetic algorithm with hybrid function (RHGA), which shows that fve
goals are overachieved, while one goal is fully achieved and one is underachieved
for Cattle-1 and 2. It can be concluded that the real parameter-based hybrid genetic
algorithm can also fnd a low-cost diet plan without violating the nutrient require-
ments (www.scialert.net).
India has the largest livestock population in world. Livestock is one of the
country’s most important economic activities, especially in the rural areas of
country, providing income for most of the family. In dairy farming, feeding costs
account for about 70% of the total operation cost. The dairying programme has
attained considerable importance in various Five-Year Plans, and the states and
the Centre have taken up several schemes/projects for the development of this
sector, but a different diet plan is needed for different categories of dairy cows:
calculating the low-cost balanced diet requires an understanding of the nutrient
requirements of dairy cows in different conditions. As per the 19th livestock
census report, the population of cows has increased by 6.52% over the previous
census report (2007), and the total number of cows estimated in 2012 was 122.9
million. The total number of milking animals in India is 116.77 million, of which
12% of the contribution is from cattle (Babic et al., 2011). Also, as per the Basic
Animal Husbandry and Fisheries Statistics 2017, the per capita availability of
average milk in Karnataka was 291 grams per day during 2016–17, which is less
than the 12 top milk-producing states in India such as Uttar Pradesh. Karnataka
has only a 4% share in milk production for the year 2016–17. From 2012 to 2016,
the cattle population increased from 1142.62 to 1370.69 (in thousands), with an
estimated milk production of 5718.22 to 6562.15 (in thousands of kg) in which the
average yield per in-milk animal of nondescript/indigenous cows from 2012–13 to
2016–17 in Karnataka was 2.32–2.43 kg/day. Area under fodder crops increased
from 35000 hectares to 36000 hectares in 2006–07, and permanent pastures and
other grazing lands decreased from 930000 hectares to 906000 hectares from
2006–07 to 2013–14 (Basic Animal Husbandry & Fisheries Statistics, 2017).
According to past surveys, it was clear that farmers are not feeding dairy cat-
tle properly due to high feed cost and unavailability of proper feedstuffs (Garg,
2012). Therefore, it is necessary to supply a least cost balanced diet to dairy
cattle, especially during pregnancy and the milking period. Since 1991 many
researchers have studied feeding practices in which small farmers have limited
resources for feeding practices (Leng, 1991: 82). The livestock industry plays an
important role in the development of the Indian economy, as the livestock sector
in agriculture GDP increased from 13.88% to 29.20% from 1990 to 2013. The
livestock sector also amounts to 4% of the national GDP (Department of Animal
Husbandry, Annual Report, 2016–17; Angadi et al., 2016). Hence, when consider-
ing the economic importance and diffculties of Indian farmers, an improvement
in feeding practice is required which results in a least cost feed plan for dairy
cows at different hypothetical conditions. Linear programming (LP) is one of the
commonly utilized strategies pursued by many economical and non-economical
diet formulation programs, but Rehman and Romero described the limitation of
LP while formulating a low-cost diet plan. The inference in the linear program-
ming model restricts the constraints (to be fxed in RHS) and objective func-
tions to a single objective, which implies an application of the goal programming
model which consists of constraints and sets of goals, which are sometimes pri-
oritized (Gupta et al., 2013b: 414–422). The main reason to apply a GP model is
to satisfy all the constraints in terms of goals. Basically, goals can be satisfed
completely or partially, or sometimes goals cannot be met. This problem where
goals cannot be met can be handled by adding +ve and −ve deviational variables
which are defned for each goal individually. After all, the objective function of
a weighted goal programming model optimizes the sum of the total deviation
from set goals where the outcome may lead to a compromise solution between
contradictory goals (Gupta et al., 2013b: 414–422). Zoran Babic et al. applied
a goal programming method to determine an optimal blend of ingredients for
livestock feed in which a GP-based model turns out to be a useful method in
deciding the optimal livestock diet formulation (Babic, 2011). Evolutionary algo-
rithms (EAs) consist of genetic algorithms (GAs), genetic programming and their
hybrid functions (Rehman [Ph.D. thesis] 2014), and an EA highly depends upon
its operators (Koda, 2012: 1–16). Furuya et al. in 1997 used genetic algorithms
in which the ratio of ingredients has evolved. Sahman et al. used a GA to fnd a
least cost diet for livestock, which results in a good solution with few constraints
(Şahman et al., 2009: 965–974). Shilpa Jain et al. did the comparative analysis of
a real and binary coded genetic algorithm on a fuzzy time series prediction. The
authors concluded that the real coded GA runs much faster than a binary coded
GA (Shilpa et al., 2013: 299–304). In a previous case study, an LP model of dairy
cows weighing 500 kg which are pregnant at three different months (seventh,
eighth and ninth months) was formulated and solved using LP simplex, GRG
nonlinear, EA and different parameters of a real coded genetic algorithm based
on primary data. This study resulted in “no signifcance difference between tech-
niques” (p > 0.05) and concluded that RGA can be used to formulate the least
cost diet. In this case study we have extended the work to formulate the goal pro-
gramming model of dairy cows, which are pregnant at third trimester, that is, in
the seventh, eighth and ninth month, which required a balanced diet to maintain
health and to produce milk with 4% fat (Kuntal et al., 2016: 594–607). This GP
model is solved using a real coded hybrid genetic algorithm.
5.3.1 Goal Programming Approach

In accord with the nutritionist, it was decided to try the linear model developed by
(Kuntal et al., 2016) by formulating it into goal programming models. In earlier
work, a linear model for Cattle-1, Cattle-2 and Cattle-3 had been developed for
cows with a body weight of 500 kg which is pregnant at third trimester that requires
a balanced ration for body maintenance and 10 litres of milk production with 4%
fat. Three goal programming models for these cattle was formulated by considering
several goals, where all the constraints except dry matter intake (DMI) are given
priority and in which least cost is highly prioritized. In earlier work, the upper and
lower bounds for each constraint has been set by the decision maker as per the
Indian Council of Agricultural Research–ICAR 2013 and NRC 2001 standard. In
this chapter, the constraints are converted to goals; their target values on a dry mat-
ter basis to fnd the diet plan are:
1. The cost should be Rs 126.71/- for Cattle-1, Rs 131.82/- for Cattle-2 and
Rs 136.65/- for Cattle-3.
2. Total dry matter intake (DMI) will be 16.75 kg for Cattle-1, 16.89 kg for
Cattle-2 and 17.03 kg for Cattle-3.
3. Crude protein (CP) will be 1.644 kg for Cattle-1, 1.691kg for Cattle-2 and
1.738 kg for Cattle-3.
4. Total digestible nutrients (TDN) will be 8.5425 kg for Cattle-1, 8.6139 kg
for Cattle-2 and 8.6853 for Cattle-3.
5. Calcium (Ca) will be 0.1176 kg for Cattle-1, 0.1223 kg for Cattle-2 and
6. Phosphorus will be 0.04193 kg for Cattle-1 and 0.04 for Cattle-2 and 3.
7. Roughage will be 12.2858 kg for Cattle-1, 12.2076 kg for Cattle-2 and
8. Concentrates will be 4.4642 kg for Cattle-1, 4.6824 kg for Cattle-2 and

4.8805 kg for Cattle-3. This forms the GP model in which seven goals
(except DMI) are formulated as goal functions. It is not easy to meet
all the seven goals; therefore, deviational variables are introduced. The
achievement (objective) function of the GP model becomes the sum of
the square root of deviation variables, which required minimization.
This goal-programming model is solved by a real coded hybrid genetic
algorithm.
5.3.1.1 GP Model-1
p1 (dcos t + ) + p2 (dCP − ) + p3 (dTDN − ) + p4 (dCa− ) + p5 (dPh− )

2 2 2 2 2
Min ( Z ) =
+p6 (dRough− ) + p7 (dConc− )
2 2
Subjected to:
1. Goal 1(Minimize Least Cost ) : °17 − +
i=1 Ci xi + dcost − dcost = 126.71
2. Goal 2 (Maximize Crude Protein ) : °17 − +

i=1 CPi + dCP − dCP = 1.644 Kg
3. Goal 3(Maximize Total Digestible Nutrient ) : °17 −

i=1 TDN i + dTDN − dTDN
+
= 8.5425 Kg
4. Goal 4 (Maximize Calcium ) : °17 − +
i=1 Cai + dCa − dCa = 0.1176 Kg
5. Goal 5(Maximize Phosphorus) : °17 − +

i=1 Phi + d Ph − d Ph = 0.04193 Kg
6. Goal 6 (Maximize Roughages) : ° i5=1 Rough i + dRough− − dRough + = 12.22858 Kg
7. Goal 7(Maximize Concentrates) : °17 − +

i=6 Conc i + dConc − dConc = 4.4
4642 Kg
8. °17
i=1 xi = 16.75 Kg
5.3.1.2 GP Model-2

2 2 2 2 2
Min ( Z ) =
2 2
Subjected to:


+
= 8.6139 Kg

i=1 Cai + dCa − dCa = 0.1223 Kg

i=1 Phi + d Ph − d Ph = 0.04 Kg
6. Goal 6 (Maximize Roughages) : °17 − +

i=1 Rough i + dRough − dRough = 12..2076 Kg

6824 Kg
8. °17i=1 xi = 16.89 Kg
5.3.1.3 GP Model-3

2 2 2 2 2
Min ( Z ) =
2 2
Subjected to:


+
= 8.6853 Kg

i=1 Cai + dCa − dCa = 0.1207 Kg

i=1 Phi + d Ph − d Ph = 0.04 Kg
6. Goal 6 (Maximize Roughages) : ° i5=1 Rough i + dRough− − dRough + = 12.11495 Kg

8805 Kg
8. °17
i=1 xi = 17.03 Kg
where pi(i = 1,2. . .7) are positive numbers between (0,1) such that p1 > p2 > . . . p7.
5.3.2 Real Coded Genetic Algorithm with Hybrid Function

Genetic algorithm is a search-based technique which is based on evolutionary
theory. Binary coded GA represents decision variables by bits of zeros and
ones, whereas GA based on real number representation is called a real coded
GA (RGA). GA works on a solution space instead of a state space, where it
builds new solutions based on the existing one. We first created an initial pop-
ulation then decided on the gene representation; we chose default population
type “double vector” to represent genes. After the representation of genes, the
population undergoes three major GA operators such as selection, crossover
and mutation to create next generations. MATLAB provides “gaoptimset” to
create or modify the GA option structure. In general, MATLAB fails to pro-
vide every possible method available in literature, but it provides lot of options
to find the optimal solution. The selection procedure decides how individuals
are selected to become parents. We used a tournament selection procedure of
size 2, where individuals may be chosen more than once as a parent. Crossover
combines two parents to create new offspring for the next generation. The
crossover heuristic returns offspring because it moves from worst parents to
past best parent, for example, if P1 and P 2 are two parents where P1 has better
fitness then offspring = P 2 + 1.2 × (P1 − P 2), where 1.2 is the ratio. Mutation
decides how an algorithm makes small changes in the individual randomly
to create new mutation offspring. Mutation is an important operator from a
diversity point of view, which allows the GA to search in a broader space. We
have linear constraints and bounds; hence, an adaptive feasible mutation is
used which generates a direction that is flexible with regard to previous favor-
able or unfavorable iterations (generations). The feasible region is bounded by
the equality and inequality constraints. An appropriate step length is chosen
towards each direction so that linear constraints as well as bounds can be sat-
isfied. After specifying these genetic algorithm options for linear models, the
GA sometimes returns a local minimum instead of global minimum, that is, a
point where the objective function value is less than the nearby points but pos-
sibly greater than the distant point in solution space. Therefore, to overcome
this deficiency of GAs, we have introduced the hybrid command “fmincon”
inside the GA, in which we allow the GA to find the valley that contains the
global minimum and after last generation; it takes the last value of GA as the
initial value of fmincon to converge quickly. In addition to that, the GA can
be tested in broader space by improving the diversity of the population, and it
can be done by manipulating the initial range of the population. However, we
have rigid constraints and bounds, so we want to search the point in the speci-
fied lower and upper bounds only. Based on the GP model, we have 31 decision
variables and seven goals (1-Equality constraint). We have to find the mini-
mum cost of diet based on dry matter; hence, we set the number of variables
to 31, from which we have developed three goal-programming models with
different priorities for cows with body weight of 500 kg which are pregnant at
third trimester.
5.3.3 Results
TABLE 5.11
Least Cost, Deviation, and Constraints Value Solved by Hybrid RGA for Goal
Programming Models
Variables Feedstuffs GP Model-1 GP Model-2 GP Model-3
x1 Paddy straw 0 0 0
x2 Co-4 grass 0 0 0
x3 Maize fodder 12.2802 12.2014 12.1431
x4 Co-FS 29 sorghum 0 0 0
fodder
x5 Ragi straw 0 0 0
x6 Maize 0 0 0
x7 Soya DOC 0 0 0
x8 Copra DOC 0.0480 0 0
x9 Cotton DOC 0.6099 0.7691 0.908
x10 Wheat bran 3.7989 3.9052 3.9654
x11 Gram chunnies 0 0 0
x12 Cotton seed 0 0 0
x13 Conc. mix Type I 0 0 0
x14 Calcite 0 0 0
x15 MM 0 0.0095 0.006
x16 DCP 0.0131 0.0048 0.0074
x17 Salt 0 0 0
Deviations GP Model-1 GP Model-2 GP Model-3
dcost − 9.7909 10.5312 11.4146

dcost+ 0 0 0
dCP− 0 0.0002 0.0001
dCP+ 0 0 0
dTDN− 0 0 0
dTDN+ 1.9453 1.9789 2.012
dCa− 0.0086 0.0115 0.01
dCa+ 0 0 0
dPh− 0.0132 0.011 0.0096
dPh+ 0.0000 0 0
dRough− 0.0056 0.0062 0.0054
dRough+ 0.0000 0 0
dConc− 0.0074 0.0081 0.0071
dConc+ 0.0000 0 0
Constraints GP Model-1 GP Model-2 GP Model-3
DMI 16.75 16.89 17.03

CP 2.3524 2.395 2.4401
TDN 10.4878 10.5928 10.6973
Calcium (Ca) 0.109 0.1108 0.1107
Phosphorus (P) 0.0287 0.029 0.0304
Roughage 12.2802 12.2014 12.1431
Concentrates 4.4568 4.6743 4.8734
Least cost (z) on 116.9191 121.2922 125.2354
DM basis
Objective 0.0127 0.0132 0.0115
function value
5.3.4 Discussion
Table 5.11 shows the results obtained for all the goal-programming models. On
assigning the weights P1 (goal 1: cost), P2 (goal 2: CP), P3 (goal 3: TDN), P4 (goal
4: Ca), P5 (goal 5: P), P6 (goal 6: roughage), P7 (goal 7: concentrate) as 0.9, 0.8,
0.7, 0.6, 0.5, 0.4, 0.3 and solving the GP Model-1 using RGA with hybrid func-
tion, we obtain dcost − = 9.7909, dTDN + = 1.9453, dCa− = 0.0086, dPh− = 0.0132,
dRough− = 0.0056 , dConc− = 0.0074 and the rest of the variables dcost + , dCP + , dCP− ,
dTDN − , dCa+ , dPh+ , dRough + , dConc + as zero. We observe that goals 1, 4, 5, 6 and 7
are overachieved and goal 3 is underachieved, while goal 2 is fully achieved with-
out any deviation, obtaining minimum (Z) = 0.0127. Similarly, on assigning the
same weights P1 (goal 1: cost), P2 (goal 2: CP), P3 (goal 3: TDN), P4 (goal 4: Ca), P5
(goal 5: P), P6 (goal 6: roughage), P7 (goal 7: concentrate) as 0.9, 0.8, 0.7, 0.6, 0.5,
0.4 and 0.3 and solving the GP Model-2 using RGA with hybrid function, we obtain
dcost − = 10.5312 , dCP− = 0.0002 , dTDN + = 1.9789 , dCa− = 0.0115 , dPh− = 0.011 ,
dRough− = 0.0062 , dConc− = 0.0081 and the rest of the variables dcost + , dCP + , dTDN − ,
dCa+ , dPh+ , dRough + , dConc + as zero. Here we also observe that goals 1, 4, 5, 6 and
7 are overachieved and goal 3 is underachieved, while goal 2 is slightly over-
achieved, with dCP− = 0.0002 obtaining minimum (Z) = 0.0132. But on assigning
the same weights P1 (goal 1: cost), P2 (goal 2: CP), P3 (goal 3: TDN), P4 (goal 4: Ca),
P5 (goal 5: P), P6 (goal 6: roughage), P7 (goal 7: concentrate) as 0.9, 0.8, 0.7, 0.6,
0.5, 0.4 and 0.3 and solving the GP Model-3 using RGA with hybrid function, we
obtain dcost − = 11.4146 , dCP− = 0.0001, dTDN + = 2.012 , dCa− = 0.01,dPh− = 0.0096 ,
dRough− = 0.0054 , dConc− = 0.0071 and the rest of the variables dcost + , dCP + , dTDN − ,
dCa+ , dPh+ , dRough + , dConc + as zero. Here it is seen that goals 1, 5, 6 and 7 are over-
achieved and goal 3 is underachieved, while goals 2 and 4 are slightly overachieved,
with deviation dCP− = 0.0001 and dCa− = 0.01 obtaining minimum (Z) = 0.0115.
The obtained result may not please the nutritionist completely; hence, he has to
work on overachieved targets. The frst, third, fourth, ffth, sixth and seventh goals
are analysed, and the reason for the overachievement can be searched for in the diet
plan. The choice of the fnal solution relies on the nutritionist, with, in this case,
three different GP-models which represent the diet plan options that a nutritionist
can choose (Babic, 2011). All possibilities are not taken into consideration, as the
LP models developed in (Kuntal et al., 2016) allow introduction of additional con-
straints anytime, which results in a new set of solutions. Some constraints (if added)
can also lead to the result of “no solution”, which means that extra additional con-
straints are so complex that it is necessary to mediate the model by increasing some
of the requirements. However, for better output, we need a further discussion with
a qualifed cattle nutritionist
5.3.5 Conclusion
The study focused on improving the results of the LP model developed by (Kuntal
et al., 2016), formulating it into goal programming models. The GP models are
solved by a real coded genetic algorithm with hybrid function to improve the qual-
ity of feed mix to the dairy cows, where this method confrms it is a useful approach
in fnding the low-cost diet plan for dairy cows at different body conditions. As the
results obtained reveal, that RGA with hybrid function can be applied to formulate
least cost rations; however, fxing the constraints and use of code for making soft-
ware is considered while choosing the technique for making the least cost diet plan.
6
Study of Real Coded Hybrid Genetic
Algorithm to Find Least Cost Ration
for Non-Pregnant Dairy Buffaloes
CONTENTS
6.1 Introduction .................................................................................................. 69
6.2 Data Collection and Data Handling............................................................. 72
6.3 Problem Description .................................................................................... 73
6.3.1 Linear Programming Model............................................................ 74
6.3.2 Goal Programming Model-1 ........................................................... 75
6.3.3 Goal Programming Model-2 ........................................................... 76
6.3.4 Goal Programming Model-3 (Auxiliary Approach) ....................... 76
6.4 Development of Real Coded Hybrid Genetic Algorithm ............................ 77
6.4.1 Objective Function, Decision Variables, Representation
of Plots ............................................................................................. 77
6.4.2 Crossover, Selection, Mutation and Elitism Operators ................... 78
6.4.3 Global vs. Local Minima................................................................. 78
6.5 Results on Dry Matter and As-Fresh Basis.................................................. 79
6.6 Discussion .................................................................................................... 80
6.7 Conclusion .................................................................................................... 86
6.1 Introduction
In the Mandya district of Karnataka, the cost of milk per litre was more in the case
of buffaloes compared to local cows due to the high fat content and high nutritive
value of buffalo milk compared to its counterpart. Based on earlier research, it was
clear that the productivity of the buffaloes maintained by different dairy farms
was lower. Therefore, there is a need to focus on two important aspects of dairy
farming: one to increase the milk productivity of buffaloes and the other one to
reduce the diet cost by upgrading scientifc dairy farming practices. Even though
there are many predefned techniques used for animal diet formulation, a strong
application of soft computing is always preferred to enhance the quality of the
solution in which the rigidity in the functions as well as constraints can be easily
handled by a linear programming program (LPP). Therefore, to meet the nutrient
requirements at the lowest cost, we have developed a real coded hybrid genetic
algorithm (RHGA) for formulating the least cost ration for dairy buffaloes, tak-
ing the hybridization of RGA and the “fmincon” of Excel Solver. This technique
is better than old conventional techniques in the sense that it does not break if
the inputs are modifed and provides better results over complex problems even
DOI: 10.1201/9781003231714-6 69
if it is a linear programming model. The linear programming model is developed

from primary data collected from the National Institute of Animal Nutrition and
Physiology (NIANP) and as per the standards of the Indian Council of Agricultural
Research (ICAR). The developed algorithms are performed equally and compared
with other low-cost diet formulation techniques in non-pregnant dairy buffalo
weighing 450 kg and producing 10 L of milk with 6% fat content as an LP model
where standard nutrient requirements are used on a dry matter basis. Further, a goal
programming model (GP model) has been developed, as there are two high priority
objectives (out of eight goals) – least cost and dry matter intake – to be achieved
simultaneously, if possible. This GP model is also solved by hybrid RGA, showing
that four goals out of eight are fully achieved. This reveals that the real parameter-
based hybrid genetic algorithm (RHGA) can be used to cut down the total mixed
ration (TMR) cost without compromising the nutrient values in the diet. Although
the GP model mentioned showed that 4 out of 8 goals are fully achieved, there was
a concern raised about the consideration of the lower and upper bounds of the goals.
Hence, a thorough review of the model in consultation with the research scientist
and nutritionists from NIANP was done, and it was concluded that 7 out of 8 goals
were of a maximizing nature (except least cost) within the permissible bounds, and
hence the negative deviations for these seven goals should be considered in the
objective function of the GP model. Only the frst goal is of a minimizing nature,
as it represents the cost function. Thus, a new GP model was developed for non-
pregnant dairy buffalo weighing 450 kg, yielding 10 L milk with 6% of fat content
that considers the standard nutrient requirement on a dry matter basis with 7 out of
8 goals as a maximization function. This model is also solved by real coded hybrid
GA (RHGA), and the results obtained are in sync with reality.
In dairy farming, feeding costs account for about 70% of the total operation costs.
Therefore, a different diet plan is required for different categories of buffaloes, in
which the most critical part is to formulate an accurate low-cost balanced diet.
While formulating the ration it is necessary to understand the nutrient requirements
of buffaloes at various stages. As per the 19th Livestock Census Report of 2003, out
of total livestock in the country, buffaloes are 21.2%; from 1997 to 2003, the buffalo
population increased by 8.9%. As per the Ministry of Agriculture Department of
Animal Husbandry, Dairying and Fisheries: Government of India, Krishi Bhavan,
19 Livestock Census–2012 All-India Report, the female buffalo population has
increased by 7.99% over the previous census, and the total number of female buffa-
loes was 92.5 million in 2012 (www.dahd.nic.in). The buffalo population increased
from 105.3 million to 108.7 million, showing a growth of 3.19%. Also, the number
of milking buffaloes increased from 111.09 million to 118.59 million, an increase
of 6.75% (19 Livestock Census–2012). This resulted in most of the farmers not
feeding their buffaloes properly due to high feed cost and unavailability of proper
feedstuffs in their respective regions, which affects the productivity of milk yield
(Garg, 2012). Most of the fbrous feed available to buffaloes is defcient in essential
nutrients like protein. The quantity as well as quality of feedstuffs decreases in the
dry season, where farmers face a major diffculty providing a proper low-cost bal-
anced ration to their animals in milk. Therefore, it is necessary that such lactating
buffaloes should be supplemented with a balanced ration which is nutritionally rich
Study of RHGA to Find Least Cost Ration 71
and should be of low cost. Since 1991, many researchers have studied the feeding
practices of dairy cattle, in which small farmers have limited resources for feed-
ing and do not have proper knowledge or resources to provide a low-cost balanced
ration (Leng, 1991). Study also reveals that in most of the country, animal diet is
imbalanced by over-feeding energy, protein and minerals. Hence, there is a large
scope for improvement in the nutritional status by adopting improved feeding prac-
tices (Mudgal et al., 2003). In addition, the dairy sector in India is highly dependent
upon buffalo milk because of its contribution to total milk production (49%), as it
is known that buffalo milk is rich in fat and solid content. The livestock industry
plays a major role in the development of the Indian economy as the share of the
livestock sector in agricultural GDP increased from 13.88% in 1980–81 to 29.20%
in 2012–13 (Reddy et al., 2016) According to the Annual Report of the Department
of Animal Husbandry, Dairying & Fisheries Ministry of Agriculture, Government
of India (2015–16), livestock is one of the most important contributors in the Indian
economy. As per the annual Report of Dept. of Animal Husbandry, Dairying and
Fisheries (2016–17) India stands frst in the world in milk production (155.5 million
tonnes). Livestock also contribute to 4% of the national GDP (Angadi et al., 2016).
A feld study on the effects of feeding a balanced ration on milk production con-
cerning nitrogen use effciency (NUE) was conducted on 7090 lactating cows and
4534 lactating buffaloes, in which using local feed resources, a balanced least cost
ration can be obtained (Garg et al., 2016). Therefore, by considering the economic
importance and diffculties of Indian farmers and the importance of buffalo rear-
ing, an improvement in feeding practices is required which results in effective milk
production by optimizing the utilization of available feedstuffs and also attempts to
minimize the feed cost of rations. Linear programming (LP) is a standout amongst
the most normally utilized strategies followed in numerous business and non-busi-
ness bolster plan programs. However, Rehman and Romero (1984) call attention to
the fact that LP has numerous restrictions while detailing apportions, practically
speaking. The presumptions in the LP strategy require objectives to be single and
requirements to be settled on the right-hand side (RHS) (Gupta et al., 2013b). This
implies the diminishment of a goal programming (GP) model comprised of limi-
tations and a set of goals which are organized a few times. The goal of GPs is to
discover the arrangement which fulfls the limitations and approach the expressed
objectives of separate issues. Hypothetically, goals could be fulflled totally,
incompletely or, in some exceptional cases, not at all. This possibility is estimated
utilizing positive and negative deviation factors that are characterized for every
objective independently, normally known as fnished or under-accomplishment
of the objective. Since the target capacity of the GP plan limits the whole of the
aggregate deviation from set objectives, the acquired outcome may yield a trade-
off arrangement between opposing objectives (Gupta et al., 2013a). Evolutionary
algorithms consist of genetic algorithm, genetic programming and their hybrid
functions (Rehman [Ph.D. Thesis] 2014), and EA highly depends upon its operators
(Koda, 2012). Furuya et al. in 1997 used genetic algorithms in which the ratio of
ingredients evolved. In this paper, he used a new form of crossover and mutation
in which mutation was generated by a combination of uniform random number and
normal distribution random numbers, and the mutation rate was very high. He also
solved the nonlinear constraint which showed that EA is good technique for a diet
problem (Tozer and Stokes, 2001). Sahman et al. (2009) used GA to fnd least cost
diet for a livestock, which results in good solution with few constraints.
In a nutshell we can say that appropriate feeding is the most important feature
in dairy buffalo management, as the feed costs account for more than half of the
total milk production (Clark and Davis, 1980). The quantity of feed that is utilized
for milk production should be optimum as well as satisfy the nutrient requirement
for the animal, to increase the proftability of dairy farmer. An LP model develops
results with a single objective of minimizing the feed cost and rigid constraints,
while converting an LP model to a GP model gives a lot of fexibility in decision
making. The rigid constraints can be considered as a goal with a specifc target with
permissible deviations, and the solution can be seen in terms of underachievement,
overachievement or fully achieved. Fixing the priorities of each goal also leads to
multiple solutions, and a judicial decision can be made depending upon the prefer-
ence of the dairy farmers. Since many researchers in the past addressed the issue
of diet problems using EA, in this chapter, we made an effort to formulate a low-
cost ration for non-pregnant buffaloes of body weight 450 kg at the third lactating
period, with 10 litre milk production containing 6% fat, which was solved using
an RHGA. The study was further extended to formulate nonlinear weighted goal
programming models by considering the objective as the square root of the sum of
the squares of the deviations.
6.2 Data Collection and Data Handling

Rations have to be formulated and need to be updated on a regular basis to avoid
overfeeding. The common guidelines for diet formulation are the Indian Council
of Agricultural Research (2013) and the National Research Council (2001).
These provide the nutrient requirements for cow and buffaloes at different con-
ditions (Table 6.1). An Excel-based computer program developed by NIANP in
Bangalore provides actual nutrient requirement values for dry matter intake for
optimizing the cost. It is diffcult to optimize the cost while considering buffaloes
at different conditions, so we obtained help from a qualifed nutritionist while
formulating the diet. A balanced diet requires frequent analyses of feedstuffs and
their cost. Table 6.2 gives the selected feedstuffs for diet formulation with their
nutrient values. Some variations in milk production can happen due to feeding
TABLE 6.1
Nutrient Requirements for Non-Pregnant Buffalo Weighing 450 kg and Yielding
10 Litres of Milk with 6% Fat
Total DMI (kg) CP (kg) TDN (kg) Ca (kg) Phosphorus (kg)
min–max min–max min–max min–max min–max
16.42–17.24 1.7158–1.8016 7.9857–9.1835 0.0680–0.0748 0.0270–0.0405
TABLE 6.2
Nutrient Content of Feedstuffs on DM Basis
DM CP TDN Ca P Cost Ci
Feedstuffs (in %) (in kg) (in kg) (in kg) (in kg) (in Rs)
Paddy straw 90 0.0513 0.45 0.0018 0.0008 5

CO-4 grass 20 0.08 0.52 0.0038 0.0036 3
Maize fodder 20 0.1086 0.58 0.0053 0.0014 3
Co-FS 29 sorghum 90 0.0823 0.52 0.003 0.0025 3
Ragi straw 90 0.06 0.42 0.0058 0.0025 3.5
Berseem 20 0.158 0.66 0.0144 0.0014 2
Wheat straw 90 0.033 0.55 0.003 0.0006 2
Maize stover 90 0.048 0.58 0.0053 0.0014 1.5
Maize 90 0.09 0.85 0.0053 0.0041 17
Soya DOC 90 0.46 0.7 0.0036 0.01 38
Copra DOC 90 0.27 0.7 0.002 0.009 23
Cotton DOC 90 0.35 0.7 0.0031 0.0072 23
Wheat bran 75 0.16 0.75 0.01067 0.00093 17
Gram chunnies 90 0.1645 0.7 0.0028 0.0054 14
Cotton seed 90 0.17 1.1 0.003 0.0062 21
Chickpea husk 90 0.16 0.65 0.004 0.0141 10
Conc. mix Type I 90 0.22 0.7 0.005 0.0045 17
Calcite 97 0 0 0.36 0 4
MM 90 0 0 0.32 0.06 50
DCP 90 0 0 0.24 0.16 28
Salt 90 0 0 0 0 5
practices, availability of feed resources or animal grouping factors. Roughages

and concentrates which are rich in digestibility should be used in the correct
proportion depending upon the lactation period and the quantity of milk with a
high percentage of fat, so depending upon conditions the roughage–concentrate
ration can vary. To obtain the least cost feed we have some constraints for each
nutrient, and it is unique for each animal. The minimum and maximum levels of
crude protein (CP 10.4–11%), total digestible nutrient (TDN 48.6–51.1%), calcium
(0.4–0.5%) and phosphorus (0.16–0.18%) can be met through different propor-
tions of roughage (40–80%) and concentrates (20–70%) on total dry matter intake
for milking buffalo.
6.3 Problem Description

The present study computes the balanced least cost ration for buffalo of body weight
450 kg in the third lactation period, where buffalo need a ration for body main-
tenance with 10 litres of milk production (6% fat content). The nutrient require-
ments of buffalo are calculated by an Excel-based computer program developed
Enter Animal Data Body Weight (in kg)

Milk Yield (in Lit)
Pregnancy
Set Feed Resources
Dry Matter Intake
Crude Protein (CP)
Total Digestible Nutrient (TDN)
Nutrient Requirement Calcium (Ca)
Phosphorus (P)
Concentrates
Least cost Feed Formulation Roughage
LP Simplex Method Real Coded GA Hybrid GA
Results on Dry Matter Basis Converted to Fresh Basis
FIGURE 6.1 Methodology to Calculate Least Cost Balanced Ration for Non-Pregnant Dairy
Buffalo; Kuntal, Adv in Intelligent System & Computing, 2017.
by NIANP, Bangalore, as per the ICAR 2013 and NRC 2001 standard. A linear
programming model is introduced for the specifed condition of buffalo. Although
there is a well-defned method to solve LPP, an evolutionary algorithm (EA) is
preferable, as these algorithms effciently deal with the mathematical rigidity of the
constraints, can handle large dimension problems and also give multiple solutions.
Since the number of variables represented as feedstuffs are greater in number and
the constraints are very rigid, we opted to develop an RHGA to solve the ration
formulation problem for buffaloes. Various versions of this algorithm were created,
and a comparative study was planned to obtain the appropriate method which can
be used by the farmers at farm level.
6.3.1 Linear Programming Model
Objective Function: Min(z) = ˜ 21

i=1 C i x i
Subjected to:
21
1. 16.42Kg ° ˜ i=1 x i °17.42Kg
21
2. 1.7158Kg ° ˜ i=1 CPi °1.8016Kg
21
3. 7.9857Kg ° ˜ i=1 TDN i ° 9.1835Kg
21
4. 0.0680Kg ° ˜ i=1 Ca i ° 0.0748Kg
21
5. 0.0270Kg ° ˜ i=1 Ph i ° 0.0405Kg
6. 6.568Kg ° ˜8i=1 Rough i °13.136Kg
21
7. 3.224Kg ° ˜ i=9 Conc i °11.494Kg
Search Space:
0.806 ° x1 ° 4.03, 0.806 ° x2 ° 4.03, 0.806 ° x3 ° 4.03, 0.806 ° x4 ° 4.03, 0.806
8
° x5 ° 4.03, 0.806 ° x6 ° 3.224, 0.806 ° x7 ° 3.224, 0.806 ° x8 ° 4.03,
0.806 ° x9 ° 3.224, 0 ° x10 ° 4.03, 0 ° x11 ° 4.03, 0 ° x12 ° 3.2240,
0 ° x13 ° 3.224, 0.1612 ° x14 ° 0.806, 0.1612 ° x15 ° 0.806, 0.0806 ° x16 ° 0.806,
.
0.806 ° x17 ° 3.224, 0 ° x18 ° 0.16120, 0 ° x19 ° 0.0806, 0 ° x20 ° 0.03224,
0.12896 ≤ x21 ≤ 0.1612
6.3.2 Goal Programming Model-1
p1 (dcost + ) + p2 (dDM + ) + p3 (dCP + ) + p4 (dTDN + ) + p5 (dCa + )

2 2 2 2 2
Min ( Z ) =
+p6 (dPh + ) + p7 (dRough + ) + p8 (dConc + )
2 2 2
Subjected to:
1. Goal 1(minimize Least Cost ) : ˜ i=1

21
Ci xi + dcost − − dcost + = 101.6073
2. Goal 2 (minimize Dry Matter ) : ˜ i=1
21
xi + dDM − − dDM + = 16.42 Kg
3. Goal 3(minimize Crude Protein ) : ˜ i=1
21
CPi + dCP − − dCP + = 1.7158 Kg
4. Goal 4 (minimize Total Digestible Nutrient ) : ˜ i=1
21
TDN i + dTDN − − dTDN
D
+
= 9.1835 Kg
5. Goal 5(minimize Calcium ) : ˜ i=1
21
Cai + dCa− − dCa + = 0.0748 Kg
6. Goal 6 (minimize Phosphorus) : ˜ i=1
21
Phi + dPh− − dPh + = 0.0405 Kg
7. Goal 7(minimize Roughages) : ˜ i=1
21
Roughi + dRough− − dRough +
= 13.13
36 Kg
8. Goal 8(minimize Concentrates) : ˜ i=1
21
Conci + dConc− − dConc + = 3.284
40 Kg
TABLE 6.3
Priority Values for GP Model-1
Priorities Goals Value
p1 Min least cost 0.9

p2 Min DM 0.8
p3 Min CP 0.7
p4 Min TDN 0.6
p5 Min Ca 0.5
p6 Min P 0.4
p7 Min roughage 0.3
p8 Min concentrate 0.2
TABLE 6.4
Priority Values for GP Model-2
Priorities Goals Value
p1 Min least cost 0.9

p2 Min DM 0.8
6.3.3 Goal Programming Model-2
Min ( Z ) = p1 (dcost + ) + p2 (dDM + )

2 2
Subjected to:
1. Goal 1(minimize Least Cost ) : ˜i=1

21
Ci xi + dcost − − dcost + = 101.6073
2. Goal 2 (minimize Dry Matter ) : ˜ i=1
21
xi + dDM − − dDM + = 16.42 Kg
6.3.4 Goal Programming Model-3 (Auxiliary Approach)
p1 (dcost + ) + p2 (dCP − ) + p3 (dTDN − ) + p4 (dCa− ) + p5 (dPh− )

2 2 2 2 2
Min ( Z ) =
2 2
Subjected to:
1. Goal 1(Minimize Least Cost ) : ˜i=1

21
Ci xi + dcost − − dcost + = 101.6073
2. Goal 2 (Maximize Crude Protein ) : ˜i=1

21
CPi + dCP− − dCP + = 1.7158 Kg
3. Goal 3(Maximize Total Digestible Nutrient ) : ˜i=1

21
TDNi + dTDN −
+
−dTDN
D = 9.1835 Kg
4. Goal 4 (Maximize Calcium ) : ˜i=1

21
Cai + dCa− − dCa + = 0.0748 Kg
5. Goal 5(Maximize Phosphorus) : ˜i=1

21
Phi + dPh− − dPh+ = 0.0405 Kg
6. Goal 6 (Maximize Roughages) : ˜8i=1 Roughi + d Rough− − d Rough +

= 13.136 Kg
7. Goal 7(Maximize Concentrates) : ˜i=1

21
Conci + dConc− − dConc+
= 3.284
40 Kg
8. ˜i21
=1 xi = 16.42 Kg
where pi (I = 1,2. . .7) are positive numbers between (0,1) such that p1 > p2 > . . . p7
6.4 Development of Real Coded Hybrid Genetic Algorithm

A genetic algorithm (GA) is a heuristic based technique which is based on evo-
lutionary theory. GA works on a solution space instead of a state space, where it
builds new solutions based on existing ones. GA combines good portions of the
solution just as nature does by combining the DNA of living beings. To use a GA,
frst we create an initial population, then we decide the “gene” representation; we
chose default population type “double vector” to represent genes due to its fex-
ibility. After selecting the representation of genes, the population undergoes three
main operators of GA such as selection, crossover and mutation operators to create
the next generation from current generations. In brief, a selection operator selects
parents that combine to populate the next generation. The crossover operator also
combines two parents to create new offspring for the next generation, and the muta-
tion operator randomly changes the individuals to create new offspring. MATLAB
provides “gaoptimset” to create or modify the GA option structure.
6.4.1 Objective Function, Decision Variables,

Representation of Plots
Firstly, we provide the objective function, that is, the function that calculates mini-
mum cost of ration of each member of population. In our linear model, we have
21 decision variables and 14 constraints from which we have to fnd the minimum
cost; hence, we set the number of variables to 21. We used “gaoptimset” for rep-
resentation of population. Population type specifes the type of input to the ftness
function, so we have restricted the input to population as “double vector” since we
do not have complex decision variables. We used some plot options in “gaoptimset”
such as @gaplotbestf, @gaplotdistance and @gaplotrange to see the performance

of population in each generation, where “gaplotbestf” plots the best ftness value
in each generation and “gaplotdistance” plots the average distance between indi-
viduals by taking 20 samples at every generation. It calculates the distance of each
sample and stores the choice, then calculates the average distance by using mean
square error. The performance of GA will be affected by the diversity of the ini-
tial population if the distance between populations is large, the diversity is large;
if the distance is low, the diversity is less. More trials are required to experiment
with the right amount of diversity because if the diversity is too high or low, GA
will not perform well and the solution may stick to local minimum. We have cre-
ated the population in the range of lower and upper bounds and tried to program
the GA to search in that space. Population size determines the size of population
at each generation; hence, we have selected four times the number of variables to
perform in every generation. Obviously, by increasing the population size, the GA
will perform better by searching more points and is more likely to provide a better
solution. In this case, we have rigid constraints and bounds; therefore, by increasing
the population size, it will take a long time for the GA to generate the population.
6.4.2 Crossover, Selection, Mutation and Elitism Operators

The selection operator in a GA selects best ft parents to perform crossover and
mutation. In a tournament selection, every individual is chosen randomly to per-
form in the tournament based on tournament size, and then the best ft individual is
selected as a parent. We chose tournament size as two. In a crossover operator, two
parents are combined to form offspring for the next iterations. A heuristic crossover
is chosen, as offspring moves from worst parents to the previous best parent, for
example, if P1 and P2 are parents and P1 has better ftness then offspring = P2 +
1.2 × (P1 − P2), where 1.2 is the ratio to perform the heuristic crossover. Mutation
decides how the algorithm makes small changes in the individuals randomly to cre-
ate the mutation offspring. It is an important parameter of the algorithm because it
provides diversity that allows the GA to search in a broader space. We have rigid
linear constraints and bounds; hence, from the GA options an adaptive feasible
mutation was preferred. These mutation options are better for linear constraints
and bounds as they creates direction which is very fexible with regard to previous
favorable or unfavorable generations (repository.um.edu). A best ft region called a
feasible region is restricted by the linear equality and inequality constraints; there-
fore, to satisfy these constraints, a step length is chosen towards the direction.
6.4.3 Global vs. Local Minima

After specifying these genetic algorithm options for linear models, many times
a GA gives local minima instead of global minima (rad.inu.edu.gr). Therefore,
to overcome this defciency of the GA we have introduced the hybrid command
“fmincon” inside the GA, in which we allow GA to fnd the valley that contains
the global minimum, and after the last generation, it takes the last value of GA as
the initial value of “fmincon” to converge quickly. Also, one can try to test GA
in the wider range by increasing the diversity of the initial population by manip-
ulating the initial population. However, we have rigid constraints and bounds,
so we want to search the point in the specifed lower and upper bounds only.
“Fmincon” is a gradient-based search technique which works on a problem with
continuous constraints and objective functions and must have a frst derivative. In
MATLAB, “fmincon” uses a sequential quadratic programming (SQP) sub-prob-
lem where SQP methods are a representation of nonlinear programming methods.
Schittkowski (Singh and Singh, 2015) has executed a version that surpasses every
other tested method in terms of effciency and accuracy of the solution for a large
number of test problems (tel.archives-ouvertes.fr). Based on previous work (Biggs,
1975; Han, 1977; Powell, 1978a), it can be seen that the method mimics Newton’s
method for a constrained problem. At every generation, an initial judgment is
made of the Hessian of the Lagrangian function using a quasi-Newton method,
which is used to generate a quadratic programming (QP) problem whose solution
is a benchmark which is used to form a search direction for a line search prob-
lem (tel.archives-ouvertes.fr). A major overview of SQP is discussed in (Fletcher,
1987; Gill et al., 1981; Powell, 1978b; Hock and Schittkowski, 1983). In light of
this, we have used the SQP algorithm for “fmincon” Solver, where the SQP algo-
rithm (identical to SQP–legacy algorithm) is the same as the active-set algorithm
but has a different implementation. An interior point algorithm can also be used,
but SQP has faster execution time and less memory usage then the SQP-legacy and
interior point algorithms. Nocedal and Wright explain the basic SQP algorithm
(Nocedal and Wright, 2006).
6.5 Results on Dry Matter and As-Fresh Basis

TABLE 6.5
Least Cost Ration Formulated by Various Techniques for Non-Pregnant Dairy Buffalo
on DM Basis (LP Model)
Minimum value obtained in 10 runs for 5000 generations
RGA with
RGA without RGA without RGA with crossover hybrid
Feedstuffs crossover mutation and mutation function
Paddy straw 3.7907 3.8114 3.187 2.363

CO-4 grass 0.8758 0.806 1.3551 0.806
Maize fodder 2.3054 2.7951 2.0736 3.086
Co-Fs 29 sorghum 1.3639 0.806 1.112 2.266
Ragi straw 1.7162 1.5121 2.0725 2.197
Berseem 1.2667 0.806 0.9759 0.806
Wheat straw 0.9611 1.601 0.8703 0.806
Maize stover 0.8572 0.806 1.3561 0.806
Maize 0.8060 0.806 0.8061 0.806
Soya DOC 0.0001 0.5647 0.0057 0
(Continued )
TABLE 6.5
(Continued)
Minimum value obtained in 10 runs for 5000 generations
RGA with
RGA without RGA without RGA with crossover hybrid
Feedstuffs crossover mutation and mutation function
Copra DOC 0.0015 0 0.0101 0

Cotton DOC 1.1233 0.4301 1.2101 0.951
Wheat bran 0.0003 0 0.0114 0
Cotton seed 0.1613 0.1612 0.1612 0.161
Chickpea husk 0.0855 0.0806 0.0911 0.081
Conc. mix Type-I 0.8076 0.806 0.8258 0.963
Calcite 0.0017 0.0117 0.0007 0
MM 0 0 0 0
DCP 0 0 0 0
Salt 0.1322 0.1612 0.129 0.161
DMI 16.419 16.42 16.419 16.42
CP 1.7148 1.7148 1.7151 1.7158
TDN 9.175 9.1835 9.1816 9.1835
Calcium (Ca) 0.0758 0.0746 0.075 0.0748
Phosphorus (P) 0.0388 0.0391 0.0412 0.0405
Roughage 13.137 12.9436 13.0025 13.1360
Concentrates 3.282 3.4765 3.4165 3.2840
Least cost (z) 104.8323 113.7639 106.0145 101.6073
6.6 Discussion
The results obtained by various techniques – viz. RGA without crossover, RGA with-
out mutation, RGA with crossover and mutation and RGA with hybrid function – for
least cost ration for dairy buffalo (which is run 10 times up to 5000 generations)
are presented in Table 6.5. The LP model for dairy buffalo has 21 variables, which
are too complex for fnding an optimal solution. RGA is also a heuristic technique,
which works on a principle of survival of best ft and tournament selection. Adaptive
feasibility mutation and heuristic crossover are used along with elitism to solve the
LP model for dairy buffalo. Although RGA without crossover, RGA without muta-
tion and RGA with crossover and mutation give a near-optimal answer, we provide
a solution using RGA with hybrid function to avoid the solution being stuck in local
minima. As farmers need a total mixed ration to feed dairy buffalo on as fresh basis,
we have converted the least cost and total mixed ration obtained by all techniques to
as fresh basis, and the results are given in Table 6.6. As per Table 6.5, this category
of animals requires about 16.42 kg of dry matter from various kinds of feeds, which
should contain 1715.8 g of protein, 9183.5 g of TDN, 74.8g of calcium and 40.5 g
of phosphorus. These nutrient requirements can be met from 13.13 kg of roughage
and 3.28 kg of concentrates on dry matter, amounting to Rupees 101.6703 per day
per cattle on dry matter basis. As fresh basis is a feed nutrient content with mois-
ture included. After converting to as fresh basis, feedstuffs required for dairy buf-
falo are approximately 36.3 kg per day, amounting to Rs 4.52 per kg using the RGA
with hybrid function. When the expected nutrient requirement solution was sought
using RGA without crossover, without mutation and with crossover and mutation,
the least cost ration obtained was Rs 4.59 per kg, Rs 4.93 per kg and Rs 4.67 per kg,
respectively. The detailed analysis of the ration is showed in Tables 6.5 and 6.6 which
exactly meets the requirements of dry matter, CP, TDN, calcium and phosphorus. The
requirements of roughage and concentrates are also met within the permissible range.
In addition, results obtained by GP Model-1, 2 and 3 are shown in Tables 6.7, 6.8 and
6.9, which reveals that from GP-1 goals 5 and 8 are fully achieved and goals 1, 2, 3, 4,
6 and 7 are underachieved. The least cost obtained by GP-1 is Rs 100.79 on DM basis,
but DMI and CP were not satisfed. Whereas by GP Model-2, our high priorities,
TABLE 6.6
Least Cost Ration Formulated by Various Techniques for Non-Pregnant Dairy Buffalo
on “As Fresh Basis” (LP Model)
RGA with RGA with
RGA without RGA without crossover and hybrid
Feedstuffs crossover mutation mutation function
Paddy straw 4.211889 4.234889 3.541111 2.625556

CO-4 grass 4.379 4.03 6.7755 4.03
Maize fodder 11.527 13.9755 10.368 15.43
Co-FS 29 sorghum 1.515444 0.895556 1.235556 2.517778
Ragi straw 1.906889 1.680111 2.302778 2.441111
Berseem 6.3335 4.03 4.8795 4.03
Wheat straw 1.067889 1.778889 0.967 0.895556
Maize stover 0.952444 0.895556 1.506778 0.895556
Maize 0.895556 0.895556 0.895667 0.895556
Soya DOC 0.000111 0.627444 0.006333 0
Copra DOC 0.001667 0 0.011222 0
Cotton DOC 1.248111 0.477889 1.344556 1.056667
Wheat bran 0.0004 0 0.0152 0
Gram chunnies 0.180444 0.505444 0.183556 0.178889
Cotton seed 0.179222 0.179111 0.179111 0.178889
Chickpea husk 0.095 0.089556 0.101222 0.09
Conc. mix Type I 0.897333 0.895556 0.917556 1.07
Calcite 0.001753 0.012062 0.000722 0
MM 0 0 0 0
DCP 0 0 0 0
Salt 0.146889 0.179111 0.143333 0.178889
TMR on as fresh basis in kg 35.5405415 35.3822285 35.3746994 36.33556
Least cost in Rs on as fresh basis 163.445699 174.679247 165.425231 164.5728
Least cost (Rs per kg) 4.59885225 4.93692045 4.67637135 4.529249
TABLE 6.7
Least Cost and Deviation Value Solved by Hybrid RGA for GP Model-1
Feedstuffs Values
On dry matter basis

Paddy straw 3.5806
CO-4 grass 0.1175
Maize fodder 0.0194
Co-FS 29 sorghum 0.1334
Ragi straw 1.0355
Berseem 2.7540
Wheat straw 2.6429
Maize stover 0.5415
Maize 1.3333
Soya DOC 1.0162
Copra DOC 0.0146
Cotton DOC 0.0240
Wheat bran 0
Gram chunnies 0.0107
Cotton seed 0
Chickpea husk 0.0214
Conc. mix Type I 0
Calcite 0
MM 0
DCP 0
Salt 0.8638
Deviations
dcost− 0.8115
dcost+ 0
dDM− 2.3113
dDM+ 0
dCP− 0.2942
dCP+ 0
dTDN− 1.5173
dTDN+ 0
dCa− 0
dCa+ 0
dPh− 0.0118
dPh+ 0
dRough− 2.3113
dRough+ 0
dConc− 0
dConc+ 0
DMI 14.1088
CP 1.4216
Feedstuffs Values
TDN 7.6663
Calcium (Ca) 0.0748
Phosphorus (P) 0.0287
Roughage 10.8248
Concentrates 3.2840
Least cost on DM basis 100.7965
As fresh basis
TMR on as fresh basis in kg 26.91883
Least cost in Rs on as fresh basis 135.0133
Least cost (in Rs per kg) 5.015569
TABLE 6.8
Feedstuffs Values
On dry matter basis

Paddy straw 1.1775
CO-4 grass 0.9571
Maize fodder 0.0001
Co-FS 29 sorghum 5.7964
Ragi straw 0
Berseem 2.5561
Wheat straw 2.6487
Maize stover 0
Maize 0
Soya DOC 0.8576
Copra DOC 0
Cotton DOC 0.0001
Wheat bran 0
Gram chunnies 0
Cotton seed 1.2705
Chickpea husk 0
Conc. mix Type I 0
Calcite 0
MM 0
DCP 0
Salt 1.1559
Deviations
dcost− 0.0001
dcost+ 0
dDM− 0
dDM+ 0
(Continued )
TABLE 6.8
(Continued)
Feedstuffs Values
dCP− 0.0032
dCP+ 0.0997
dTDN− 0.2440
dTDN+ 1.5679
dCa− 0.0055
dCa+ 0.2868
dPh− 0.0663
dPh+ 0.7203
dRough− 0.0216
dRough+ 0.0182
dConc− 0.2544
dConc+ 0.0095
DMI 16.4200
CP 1.7158
TDN 9.1835
Calcium (Ca) 0.0748
Roughage 13.1359
Concentrates 3.2841
As fresh basis
TABLE 6.9
Feedstuffs Values
On dry matter basis

Paddy straw 0.0272
CO-4 grass 0
Maize fodder 0
Co-FS 29 sorghum 0
Ragi straw 0.0171
Berseem 13.074
Wheat straw 0.0156
Maize stover 0.0022
Maize 0.0505
Soya DOC 0
Copra DOC 0.0053
Feedstuffs Values
Cotton DOC 0.0006

Wheat bran 0.0227
Gram chunnies 0.0231
Cotton seed 0.4741
Chickpea husk 0.0151
Conc. mix Type I 0.0057
Calcite 1.4125
MM 1.1399
DCP 0.0001
Salt 0.1344
Deviations
dcost− 0.0021
dcost+ 0
dCP− 0
dCP+ 0.4508
dTDN− 0
dTDN+ 0.0901
dCa− 0
dCa+ 0.9891
dPh− 0
dPh+ 0.0499
dRough− 0
dRough+ 0
dConc− 0
dConc+ 0
DMI 16.4200
CP 2.1666
TDN 9.2736
Calcium (Ca) 1.0639
Roughage 13.1360
Concentrates 3.2840
As fresh basis
goals 1 and 2, are completely satisfed and the rest of the goals are either under-
achieved or overachieved, but the least cost obtained was Rs 100.6090 per day per
cattle on dry matter basis, which is same as the LP model. The amount of feedstuffs is
much less compared to the LP model, which is truly an advantage to farmers because
of the limitation of feedstuff. From a farmer’s point of view, the TMR required for
the farmer by the RGA hybrid function is 26.9 kg per day, amounting to Rs 5.01
per kg, whereas by GP Model-2, the TMR required is 30 kg per day, amounting to
approximately Rs 4.7 per kg, which is less compared to the RGA hybrid function.
However, for a better output for farmers, we need a further discussion with a qualifed
cattle nutritionist. For GP Model-3, we observe that on assigning the weights P1 (goal
1: cost), P2 (goal 2: CP), P3 (goal 3: TDN), P4 (goal 4: Ca), P5 (goal 5: P), P6 (goal 6:
roughage), P7 (goal 7: concentrate) as 0.9, 0.8, 0.7, 0.6, 0.5, 0.4 and 0.3 and solving the
GP Model using RGA with hybrid function, we obtain dcost− = 0.0021, dCP+ = 0.4508,
dTDN+ = 0.0901, dCa+ = 0.9891, dPh+ = 0.0499, and rest of the deviational variables dcost+,
dCP−, dTDN−, dCa−, dPh−, dRough−, dRough+, dConc−, dConc+ are zero. We observe that goal 1
is overachieved, goals 2, 3, 4 and 5 are underachieved and goals 6 and 7 are fully
achieved without any deviation with minimum (Z) = 0. The results obtained in this
work are better than the results obtained for GP Model-1. On comparing the results
obtained in both the models, dcost−, dDM−, dCP−, dTDN−, dPh−, dRough− are overachieved,
while dCa−, dCa+, dConc−, dConc+ are fully achieved, due to which all constraints except
calcium and concentrates are satisfed with the least cost of Rs 100.605/- on DM
basis. The reason for not satisfying the constraints by the goal programming model is
adding the deviation variables to dry matter basis; that is, dry matter intake is one of
the fxed constraints, and hence it has to be treated as a constraint rather than treat-
ing it simply as a goal. In the present work, after considering all the loopholes, a new
goal-programming model has been developed which gives an improved solution in
which almost all the constraints are satisfed including our high priority goals (least
cost and dry matter intake). The obtained result was not completely accepted by the
nutritionist, as the reasons for underachieved and overachieved targets need to be
analysed, because the choice of a fnal solution depends on the nutritionist, and there
is a need of further discussion with the nutritionist for better output.
6.7 Conclusion
The present study addresses the use of a real coded genetic algorithm with hybrid
function (RHGA) as a tool to provide a good quality feed mix to the dairy buf-
falo for better health and milk production. All the techniques – viz. RGA with-
out crossover, RGA without Mutation, RGA with crossover and mutation and RGA
with hybrid function – for the least cost ration are performing equally. RGA without
crossover and RGA without mutation operator provide near-optimal answers, but
the solutions seem to be get stuck in local minima; hence, it is proved that the real
coded genetic algorithm with hybrid function (RHGA) provides an optimal solution,
and this method can be used effectively for a low-cost diet plan for dairy buffalo.
Nutrient requirements of buffalo are calculated by an Excel-based computer pro-
gram developed by NIANP, Bangalore, as per the Indian Council of Agricultural
Research–ICAR 2013 and NRC 2001 standard. Application of the RHGA in solv-
ing GP Model-1, 2 and 3 for the ration formulation problem of Indian buffaloes
in the Mandya district of Karnataka is found to be very effcient. Further detailed
research with various species of animals and with different physiological needs may
be required to fne-tune the techniques for farmers’ use. In addition to that, we are
able to cut down the cost of TMR without any nutrient defciency in the diet.
7
Conclusion
The main purpose of formulating the diet problem is to fnd the combination of
feed ingredients which satisfes the entire daily nutritional requirements of dairy
animals with minimum cost. As the farmer’s objective is to reduce the cost of
ration and to maximize the proft, since spending more on rations will decrease the
proftability of farmer, this can be achieved by scientifc and economic effciency.
This study is the proof that optimization of the ration at a cheaper rate is pos-
sible at different body conditions of dairy cows and buffaloes by combining linear
and goal programming (LP and GP) with priority functions and using a heuristic
technique called a real coded genetic algorithm (RGA). The entire idea was frst
tested with a nonlinear model for Sahiwal cows with the objective of maximizing
the milk yield by Pratiksha Sexena using a binary coded genetic algorithm. The
effectiveness of the binary coded genetic algorithm was also tested on three linear
programming model and a nonlinear goal programming model extracted from the
work of Shrabani Gosh, where these mathematical models are developed for differ-
ent conditions, viz: if a fully grown cow having a body weight of 500 kg:
i. Does not produce milk,

ii. Produces different levels of milk with a certain amount of fat,
iii. Is pregnant in the third trimester, where it needs extra nutrient supplements.
Later, three linear programming (LP) models and nonlinear programming (NLP)
models are formulated for the least cost diet of dairy cows of body weight 500 kg,
where the cows during pregnancy at third trimester need a balanced ration for body
maintenance, milk production (with a minimum of 4% of fat content) and for fetus
growth. These models are solved by an RGA (with and without seeding the random
number generation), and the results are compared with Excel Solver tools, namely
GRG nonlinear, evolutionary algorithm (EA) and simplex LP. Further, the study
has been extended to the development of linear and goal programming models of
non-pregnant dairy buffalo of body weight 450 kg, which need a ration to produce
10 litres of milk with 6% fat content. It has been observed that in all these case
studies, reductions of ration cost is possible by heuristic techniques as compared to
the old conventional method, irrespective of the rigidity of constraints. For better
diet formulation, the minimum and maximum range (in percentage) of crude pro-
tein (CP), total digestible nutrients (TDN), dry matter (DM), calcium, phosphorus,
roughage and concentrates are introduced in both LP and GP models based on ICAR
2013 and NRC 2001 standards. The RGA and hybrid genetic algorithm (RHGA)
DOI: 10.1201/9781003231714-7 87
88 Conclusion
were developed to overcome the limitation of binary coded GA. In Chapter 4, we

considered the case study in which Pratiksha Sexena solved a nonlinear model
using the Kuhn-Tucker method, which resulted in the values of three nutrient ingre-
dients, CP, TDM and DM as x1 = 782.97800, x2 = 67.00717 and x3 = 507.79209
gm/kg metabolic body weight, with the maximum objective value of 582.01gm/kg
metabolic weight. Results obtained for the three nutrients using the genetic algo-
rithm for the nonlinear model with the original bounds is x1 = (721.3220, 747.1494),
x2 = (71.5456, 74.6055), x3 = (432.3138, 474.6072) gm/kg metabolic body weight,
with maximum objective value of 819.1805 gm/kg metabolic weight (Gupta et al.,
2013b). For reduced bounds, the wide range of solutions obtained is x1 = (670.5081,
677.1270), x2 = (68.8769, 69.4029), x3 = (428.7671, 463.4859) gm/kg metabolic
weight, with the global maxima of 593.8254 gm/kg metabolic weight (Gupta et al.,
2013b). Further, the nonlinear model was also solved by controlled random search
technique (RST2) with original and reduced bounds, which revealed that varia-
tions in the value of decision variables x1, x2 and x3 are 4.3%, −0.19% and 20.17%,
respectively, which leads to 0.68% variation in maximum milk yield (objective
function). There is approximately 2% deviation in the objective function compared
to the Kuhn-Tucker method. Therefore, the solution of NLPP (1) and (2) using a
genetic algorithm reveals that the variation of decision variables x1, x2 and x3 are
9.3%, 6.9% and 6.08%, respectively, which leads to a 27% variation in the objective
function and is more compared to RST2 (Gupta et al., 2013b). Using a wide range
of bounds for GA, results were obtained only after 3000 generations, whereas after
reducing the bounds the best result is obtained in 500 generations. Hence, reduc-
tion of bounds gives the better result. After comparing the results obtained by GA
with RST2, it reveals that both techniques are comparable, with a small deviation
of 4.6%. Therefore, there is a possibility of solving the nonlinear model of animal
diet using GA with a slight modifcation in the technique according to the require-
ment of the planner. GA gives fexibility to choose any option from the wide range
of solutions that act as an additional bonus to the planner.
The second case study in Chapter 4 is extracted from the work of Shrabani Gosh,
which consists of three nonlinear goal programming models for Level-1, 2 and 3 to
test the effect of a binary genetic algorithm on animal diet formulation. The focus
was to discuss the results of deviational variables obtained by binary coded genetic
algorithm developed by Mitsuo Gen et al. (1997) rather than discussing the details
of model formulation. This study reveals that the GA approach gives better results
as compared to RST (see Figure 4.7). It is also noticed from the results shown in
Table 4.7 that goal 1 and goal 15 were overachieved for Level-1. In Level-2, goal
16 is overachieved, and in Level-3, goal 1 is overachieved. All other goals are fully
achieved in a thousand generations using GA for the linear model. Figure 4.5 can
be referred to see the wholistic view of the goals’ achievements. Figure 4.6 displays
the goals’ achievement levels for a nonlinear model. We can see that goal 1 and
goal 15 are overachieved for Level-1. In Level-2, all the goals are achieved, and for
Level-3, goal 1 and goal 16 are overachieved. All other goals are fully achieved in
1000 generations using GA for a nonlinear model. Therefore, it can be concluded
that by fne-tuning the nutritive goals and by introducing harmless deviations as
underachievement and overachievement, binary coded genetic algorithm can be
Conclusion 89
applied effectively for fnding the least cost diet plan. In addition, after analysing
the approach applied by Shrabani Ghosh, it was found that minerals are not used
while formulating the diet plan. But there are three different body conditions of
animals, which defnitely requires additional nutrient supplements called minerals,
and this supplement cannot be obtained from feedstuffs.
In Chapter 5, we took into account this important aspect of including minerals
as essential nutrient requirement for the animal while formulating the diet plan for
cattle in the Mandya district of Karnataka. Using the primary data of the Mandya
district of Karnataka, taken from NIANP Bangalore, three different linear mod-
els were formulated for dairy cattle with body weight 500 kg and a 10 litre milk
yield with 4% of fat content who were in the seventh, eighth and ninth months of
pregnancy. The constraints for these LP models are fxed by giving a minimum
and maximum percentage range on dry matter basis, which was not considered
by researchers earlier. This feature makes the model easy to understand, but at the
same time the model becomes rigid. The minimum and maximum level of CP is
9.82–11%, TDN is 46.73–51%, calcium is 0.38–0.8%, phosphorus is 0.23–0.5%,
roughage is 40–80% and concentrate is 20–70%, as calculated on total DMI for
each type of cattle. These LP models are solved using LP simplex, GRG nonlinear,
evolutionary algorithm (EA) and real coded genetic algorithm (RGA) with seed-
ing the random numbers. The LP simplex method provides an iterative algorithm
to locate the corner points systematically until we get an optimal solution. The
GRG nonlinear algorithm with forward differencing deals with problems involving
decision variables and nonlinear constraints. Forward differencing uses a single
point that is slightly different from the current value to fnd the derivative. Solving
LP models in Excel Solver, it will not compute derivatives again and again, and it
continues to estimate the solution along the straight line instead of recalculating the
changing gradients. When GRG fnds a solution, this means that Solver has found
a valley for minimizing the objective function after satisfying the Karush-Kuhn-
Tucker (KKT) conditions for local optimality, and there are no other possible solu-
tions for decision variables (feedstuffs) near to current values. EA is also used to
solve the LP model by seeding the random number generator. The results obtained
by various techniques for LP models, are presented in Tables 5.4, 5.6 and 5.8 for
Cattle-1, 2 and 3 on DM basis. LP models 1, 2 and 3 have 17 decision variables
and eight constraints with minimum and maximum range. In this study, the EA
method with seeding technique could not fnd least cost because costlier nutrients
like the crude protein level in fnal feed was higher (1.82 kg) in EA as compared to
other method (1.644 kg). Similarly, TDN content is also higher in the EA method
than other methods. These could be the reasons for higher costs (INR 163.14)
obtained using EA compared to other methods (GRG–126.71; LP–126.71: RGA
seeding 1–129.44 and RGA seeding 17–128.8). It is evident that the EA method
is not so accurate for least cost feed formulation because it uses only mutation as
a parameter to improve the diversity of the population in every generation. EA is
heuristic in nature but uses only mutation as one parameter to improve the diversity
of the problem. Hence, while solving the problem the constraints are not satisfed
properly and almost all the constraints are violated. If we do not seed the random
number, then we get the wide ranges of solutions in every run most of the time, but
90 Conclusion
unfortunately in this study we did not achieve the optimal solution using EA in MS
Excel Solver due to the rigidity of the constraints.
An RGA with seeding technique is developed to overcome the limitation of EA.
RGA is also a heuristic technique which works on the principle of the survival
of best ft and tournament selection. Adaptive feasibility mutation and heuristic
crossover is used along with elitism to solve the LP model for Cattle-1, 2 and 3.
Though the seeding technique gives near-optimal answers, we prefer to provide
the wide ranges of solutions to farmers in which the values of feedstuffs change in
every run where all the constraints will satisfy. As a farmer needs a total mixed
ration to feed the cattle on as fresh basis, we have converted the least cost and total
mixed ration obtained by all techniques to as fresh basis, and the results are given
in Tables 5.5, 5.7 and 5.9. As per the ftment in Table 5.4, an animal of each cat-
egory requires about 16.75 kg of dry matter from various feed ingredients, which
should contain 1644 g of protein, 8542 g of TDN, 117.6 g of calcium and 42 g of
phosphorus, and these nutrients can be met from 12.29 kg of roughages and 4.464
kg of concentrate on dry matter. The corresponding TMR cost on dry matter basis
is Rs 126.70, that is, Rs 7.56 per kg. As fresh basis is a feed nutrient content with
moisture included. After converting to as fresh basis, the feedstuff requirement for
Cattle-1 is approximately 30 kg/day, amounting to Rs 5.83/kg using the GRG non-
linear and simplex LP technique. When the expected nutrient requirements were
tested using an RGA without seeding and with seeding (RGA1 and RGA17), the
least cost ration obtained was Rs 6.05/kg, Rs 5.96/kg and Rs 5.99/kg, respectively.
The detailed analysis of rations showed (in Tables 5.4 and 5.5) that this method
exactly met the requirement of dry matter, TDN, CP, calcium and phosphorus, and
roughage–concentrate was also well within the permissible range. Similar analy-
sis has been done for Cattle-2 (see Tables 5.6 and 5.7), where the least cost ration
obtained by various techniques for as fresh basis is Rs 6.08, Rs 6.25, Rs 6.40 and
Rs 7.20 per kg by LP simplex, GRG nonlinear, RGA (1) and RGA (17), respectively.
For Cattle-3 (see Tables 5.8 and 5.9), the TMR cost for as fresh basis turns out to
be Rs 6.38, Rs 6.59, Rs 6.50, Rs 6.50 per kg, respectively. Table 5.10 shows the net
proft that farmers can make per cattle in Indian rupees for 10 litres of milk each
day on as fresh basis by all techniques. A one-way ANOVA test at 5% level of
signifcance has been performed for the “Null hypothesis: there is no signifcant
difference between techniques”. The test reveals that since the P value is greater
than 0.05, there is no signifcance difference between the techniques. Hence, it is
proved that the real coded genetic algorithm (RGA) method can be used for ration
formulation to fnd least cost feedstuffs in dairy cattle.
A further linear model was converted to formulate the goal programming model
with priority functions. By applying the GP model in this study, the cost of rations
could be reduced to a reasonable extent satisfying the exact nutrient requirement.
The cost of diet can be reduced by giving frst priority to underachievement of least
cost and the other priority to overachievement of other nutrients such as CP, TDN,
Ca, P, concentrates and roughages on a dry matter basis. Basically, many researchers
will try to adjust the dry matter level or change the ratio of concentrates–roughage
to reduce the cost of diet by using the GP model, but we have used the same range
for constraints which was used earlier for the LP model. Table 5.11 shows the result
Conclusion 91
obtained for GP Model-1, 2 and 3 by the real coded genetic algorithm after assign-
ing the weights on P1 (goal 1: cost), P2 (goal 2: CP), P3 (goal 3: TDN), P4 (goal 4:
Ca), P5 (goal 5: P), P6 (goal 6: roughage) and P7 (goal 7: concentrate) as 0.9, 0.8, 0.7,
0.6, 0.5, 0.4 and 0.3. The results obtained reveal that for GP Model-1, using RGA
with hybrid function, we obtain dcost− = 9.7909, dTDN+ = 1.9453, dCa− = 0.0086, dPh− =
0.0132, dRough− = 0.0056, dConc− = 0.0074 and the rest of the variables dcost+, dCP+,
dCP−, dTDN−, dCa+, dPh+, dRough+, dConc+ as zero. Goals-1, 4, 5, 6 and7 are overachieved
and goal-3 is underachieved, while goal-2 is fully achieved without any deviation,
obtaining minimum (Z) = 0.0127, which results in a cost of Rs 116.9191/- per kg, per
cattle. For GP Model-2, the deviational values are dCP− = 10.5312, dCP− = 0.0002,
dTDN+ = 1.9789, dCa− = 0.0115, dPh− = 0.011, dRough− = 0.0062, dConc− = 0.0081, and the
rest of the variables dcost+, dCP+, dTDN−, dCa+, dPh+, dRough+, dConc+ as zero, where goals-
1, 4, 5, 6 and7 are overachieved and goal-3 is underachieved, while goal-2 is slightly
overachieved with dCP− = 0.0002 obtaining minimum (Z) = 0.0132, amounting to
Rs 121.2922/- per kg, per cattle on DM basis. For GP Model-3, the deviational
values are dCP− = 11.4146, dCP− = 0.0001, dTDN+ = 2.012, dCa− = 0.01, dPh− = 0.0096,
dRough− = 0.0054, dConc− = 0.0071 and the rest of the variables dcost+, dCP+, dTDN−, dCa+,
dPh+, dRough+, dConc+ as zero where goals-1, 5, 6 and7 are overachieved and goal-3 is
underachieved, while goals-2 and 4 are slightly overachieved with deviation dCP− =
0.0001 and dCa− = 0.01 with minimum (Z) = 0.0115, amounting to Rs 125.2354/- per
kg, per cattle on DM basis. Overall, the cost of Cattle-1, 2 and 3 obtained by the
goal programming approach is cheaper by 7.72% for Cattle-1, 7.98% for Cattle-2
and 8.34% for Cattle-3 in comparison with the results obtained by LP model-1, 2
and 3 on dry matter basis, respectively.
The results obtained by the GP model do not completely satisfy the decision
maker; therefore, the decision maker needs to work on the overachieved targets.
First, the third, fourth, ffth, sixth and seventh goals are analysed, and the reason
for the overachievement can be searched for in the diet plan. All possibilities are not
considered, as the LP model developed earlier allows the introduction of additional
constraints anytime, which results in a new set of solutions, while some constraints
(if added) can also lead to “no solution”, which means that additional constraints
are too complex and it is necessary to mediate in the model by increasing some of
the requirements. However, for better output, we need a further discussion with a
qualifed cattle nutritionist.
Further, in Chapter 6, three goal programming models for non-pregnant buffalo
weighing 450 kg and yielding 10 litres of milk with 6% fat are developed on a dry
matter basis. In GP Model-1, all eight goals are considered for minimization. In GP
Model-2, only two goals, least cost and dry matter, are considered for minimiza-
tion, and in GP Model-3, other than least cost, all other goals are of a maximizing
nature. These GP models are solved by an RGA with the hybrid function “fmin-
con”, and the results obtained reveal that an RHGA can effectively be used to econ-
omize the total mixed ration cost such that the feed requirements of the animals
are met without any nutrient defciency. Initially, the linear model is considered and
solved by real coded GA with different parameters, and due to the large dimension
of the problem and rigidity in constraints, we opted to develop an RHGA to solve
the ration formulation problem for buffaloes. This category of animal requires 16.42 kg
92 Conclusion
of dry matter from various kinds of feedstuffs, which should contain 1715.8 g of
protein, 9183.5 g of TDN, 74.8 g of calcium and 40.5 g of phosphorus. As per the
LP model solved using hybrid GA, this requirement can be met from 13.13 kg of
roughage and 3.28 kg of concentrates on dry matter. Its corresponding cost on dry
matter basis is Rs 101.6703/- per kg, per buffalo. As fresh basis is a feed nutrient
content with moisture included for total mixed ration (TMR); hence, it requires
approximately 36.3 kg TMR, amounting to Rs 4.5/- per kg. By GP Model-1 the
approximately cost of ration is Rs 100.7965/-on DM basis, which corresponds to
26.9 kg TMR, amounting to Rs 5.01/- per Kg, where the goals 1, 2 3, 4, 6 and 7 are
overachieved and goals 5 and 7 are fully achieved. This GP model corresponds to
least cost, but the constraints such as least cost, CP, TDN, P, roughage and concen-
trates are decreased by 14%, 17.4%, 16.5%, 29.13% and 17.594%, and calcium and
phosphorus are achieved completely, 100%. Therefore, to overcome this defciency,
GP Model-2 is formulated with only two priority goals (minimizing least cost and
DMI), and the results obtained revealed that all the goals are completely achieved
with least cost of Rs 101.6090/- on DM basis, which leads to approximately 30 kg of
TMR, amounting to Rs 4.7/- per kg. Goals 1 and 2 are fully achieved, and remain-
ing goals are either underachieved or overachieved. GP Model-3 is also formulated
with the objective of minimizing the cost of the ration by not considering dry mat-
ter as a goal in the objective function. Dry matter is treated as a constraint directly,
from which we observe that on assigning the weights P1 (goal 1: cost), P2 (goal
2: CP), P3 (goal 3: TDN), P4 (goal 4: Ca), P5 (goal 5: P), P6 (goal 6: roughage), P7
(goal 7: concentrate) as 0.9, 0.8, 0.7, 0.6, 0.5, 0.4 and 0.3 and solving the GP Model
using RGA with hybrid function, we obtain: dcost− = 0.0021, dCP+ = 0.4508, dTDN+ =
0.0901, dCa+ = 0.9891, dPh+ = 0.0499, and the rest of the deviational variables dcost+,
dCP−, dTDN−, dCa−, dPh−, dRough−, dRough+, dConc−, dConc+ are zero. We observe that goal 1
is overachieved; goals 2, 3, 4 and 5 are underachieved, while goals 6 and 7 are fully
achieved without any deviation, with a minimum cost of Rs 101.605/- on DM basis.
All three GP models give a wide range of solutions, but the obtained result does not
completely satisfy the decision maker/nutritionist, as the reasons for underachieve-
ment and overachievement targets need to be analysed further, where the choice of
a fnal solution depends upon the decision maker. There is a need for further discus-
sion with a nutritionist for better output.
The results of the study from Chapters 4, 5 and 6 reveal that after keeping the
nutritional requirements of dairy cattle and buffalo at different body conditions,
the integrated linear and simple goal programming model with weighted priority
function was suffcient to optimize the ration cost. The ration cost obtained by this
method was reasonably cheaper than when formulated by only the LP approach.
The requirement of nutrients (CP, TDN, Ca, P, roughage and concentrates) was
met by fne-tuning the individual feedstuffs. The roughage–concentrates and other
constraints maximum–minimum range make the model more rigid, and hence the
binary coded genetic algorithm was not suggested to solve the problem. To over-
come the limitation of the binary genetic algorithm, it was also proved that the
use of the RGA is a better technique for a good quality feed mix to dairy cattle
for better health and milk production over other conventional methods, viz. LP
simplex, GRG nonlinear and EA and hence can also be used for ration formulation
Conclusion 93
to fnd least cost feedstuffs in dairy cattle. All the techniques – viz. RGA without
crossover, RGA without mutation, RGA with crossover and mutation and RGA with
hybrid function – for least cost ration perform equally. RGA without crossover and
RGA without mutation provide near-optimal answers, but solutions seem to get
stuck in local minima; hence, it is proved that RGA with a hybrid function provides
the optimal solution, and this method can also be used for a ration formulation to
fnd least cost feedstuffs for dairy buffalo.
References
19 Livestock Census–2012. All India Report Ministry of Agriculture Department of

Animal Husbandry. New Delhi: Dairying and Fisheries Krishi Bhavan.
Afolayan, M., Olatunde, A. M. and Afolayan, M. 2008. ‘Nigeria Oriented Poultry Feed
Formulation Software Requirements’. Journal of Applied Sciences Research. Vol. 4.
Pp. 1596–1602.
Afrouziyeh, M., Shivazad, M., Chamani, M. and Dashti, G. 2010. ‘Use of Non-linear
Programming to Determine the Economically Optimal Energy Density in Laying
Hens’ Diet During Phase 1’. African Journal of Agricultural Research. Vol. 5.
Pp. 2270–2777.
Al-Deseit, B. 2009. ‘Least-Cost Broiler Ration Formulation Using Linear Programming
Technique’. Journal of Animal and Veterinary Advances. Vol. 8. Pp. 1274–1278.
Aletor, V. A. 1986. ‘Some Agro-Industrial By-products and Wastes in Livestock Feeding:
A Review of Prospects and Problems’. World Review of Animal Production. Vol. 22.
Pp. 35–41.
Ali, M. A. and Leeson, S. 1995. ‘The Nutritive Value of Some Indigenous Asian Poultry
Feed Ingredients’. Animal Feed Science and Technology. Vol. 55. Pp. 227–237.
Almasad, M., Altahat, A. and Al-Sharafat, A. 2011. ‘Applying Linear Programming
Technique to Formulate Least Cost Balanced Ration for White Egg Layers in
Jordan’. International Journal of Empirical Research. Vol. 1. Pp. 112–120.
Angadi, U. B., Anandan, S., Gowda, N. K. S., Rajendran, D., Devi, L., Elangovan, A.
V. and Jash, S. 2016. ‘“Feed Assis” – An Expert System on Balanced Feeding for
Dairy Animals’. AGRIS On-line Papers in Economics and Informatics. Vol. 8(3).
Pp. 3–12.
Annual Report. 2016. Department of Animal Husbandry, Dairying and Fisheries,
Ministry of Agriculture and Farmers Welfare. Govt. of India.
Babic, Z. and Peric, T. 2011. ‘Optimization of Livestock Feed Blend by Use of Goal
Programming’. International Journal of Production Economics. Vol. 130.
Pp. 218–223.
Bajpai, P. and Kumar, M. 2010. ‘Genetic Algorithm—an Approach to Solve Global
Optimization Problems’. Indian Journal of Computer Science and Engineering.
Vol. 1(3). Pp. 199–206.
Baron, L., Achiche, S. and Balazinski, M. 2002. ‘Fuzzy Decisions Systemknowledge
Base Generation Using a Genetic Algorithm’. International Journal of
Approximate Reasoning. Vol. 28. Pp. 125–148.
Basic Animal Husbandry & Fisheries Statistics. 2017. AHS Series-18. New Delhi:
Government of India Ministry of Agriculture & Farmers Welfare Department of
Animal Husbandry. Dairying and Fisheries Krishi Bhavan.
Bath, D. L. 1985. ‘Nutritional Requirements and Economics of Lowering Feed Costs’.
Journal of Dairy Science. Vol. 68(6). Pp. 1579–1584.
Bharti. 1993. ‘Controlled Random Search Optimization Techniques and Their
Applications’. Ph.D. Thesis. India: U.O.R. Roorkee.
95
96 References
Biggs, M. C. 1975. ‘Constrained Minimization Using Recursive Quadratic Programming’.

In Towards Global Optimization (L. C. W. Dixon and G. P. Szergo, eds.). North
Holland. Pp. 341–349.
Black, J. R. and Hlubik, J. 1980. ‘Basics of Computerized Linear Programs for Ration
Formulation’. Journal of Dairy Science. Vol. 63. Pp. 1366–1377.
Bremermann, H. J. and Anderson, R. W. 1989. ‘An Alternative to Backpropagation:
A Simple Rule of Synaptic Modifcation for Neural Net Training and Memory’.
Technical Report: U.C. Berkeley Center for Pure and Applied Mathematics
PAM-483.
Brooks, S. H. 1958. ‘A Discussion of Random Methods for Seeking Maxima.’ Operations
Research. Vol. 6. Pp. 244–251.
Broyden, C. G. 1967. ‘Quasi-Newton Methods and Their Application to Function
Minimization’. Mathemutics of Computation. Vol. 21. Pp. 368–381.
Castrodeza, C., Lara, P. and Pen, T. 2005. ‘Multicriteria Fractional Model for Feed
Formulation: Economic, Nutritional and Environmental Criteria’. Agricultural
Systems. Vol. 86. Pp. 76–96.
Chakeredza, S., Akinnifesi, F. K., Ajayi, O. C., Sileshi, G., Mngomba, S. and Gondwe,
F. M. T. 2008. ‘A Simple Method of Formulating Least-cost Diets for Smallholder
Dairy Production in Sub-sub-Saharan Africa’. African Journal of Biotechnology.
Vol. 7. Pp. 2925–2933.
Chavas, J.-P., Kliebenstein, J. and Crenshaw, T. ‘Modeling Dynarnic (August 1985):
Agricultural Production Response: The Case of Swine Production.’ American
Journal of Agricultural Economics. Vol. 67. Pp. 636–646.
Clark, J. H. and Davis, C. L. 1980. ‘Some Aspects of Feeding High Producing Dairy
Cows’. Journal of Dairy Science. Vol. 63. P. 873.
Dantzig, G. B. 1951. ‘Application of the Simplex Method to a Transportation Problem’,
Activity Analysis of Production and Allocation. Koopmans, T.C., Ed., John Wiley
and Sons, New York, Pp. 359–373.
Dantzig, G. B. 1963. Linear Programming and Extension. Princeton, NJ: Princeton
University Press and the Rand Corporation.
Davies, D. 1970. ‘Some Practical Methods of Optimization’. In Integer and Nonlinear
Programming (J. Abadie ed.), North-Holland Publishing Co., Amsterdam, 1970.
Davis, L. 1989. ‘Adapting Operator Probabilities in Genetic Algorithms’. In Proceedings
of the Third International Conference on Genetic Algorithms (J. David Schaer
ed.). San Mateo: Morgan Kaufmann Publishers). Pp. 61–69.
Deb, K. 1999. ‘An Introduction to Genetic Algorithms’. Sadhana. Vol. 24. Pp. 293–315.
Doi: https://doi.org/10.1007/BF02823145.
Deb, K. 2009. Optimization for Engineering Design; 2010, PHI. ISBN-978-81-203-0943-2.
Dent, J. B. and Casey, H. 1967. Linear Programming and Animal Nutrition. London:
Crosby Lockwood.
Dixon, L. C. W. 1972a. Nonlinear Optimization. London: English Universities Press.
Dixon, L. C. W. 1972b. ‘Variable Metric Algorithms: Necessary and Suffcient Conditions
for Identical Behavior of Nonquadratic Functions’. Journal of Optimization
Theory and Applications. Vol. 10. Pp. 34–40.
Dixon, L. C. W. 1973a. ‘ACSIM: An Accelerated Constrained Simplex Technique’.
ComputerAided Design. Vol. 5. Pp. 23–32.
References 97
Dixon, L. C. W. 1973b. ‘Conjugate Directions Without Line Searches’. Journal of the

Institute of Mathematics Applications. Vol. 11. Pp. 317–328.
Dixon, L. C. W. (ed.). 1976. Optimization in Action. New York: Academic Press.
Eila, N., Lavvaf, A. and Farahvash, T. 2012. ‘A Model for Obtaining More Economic
Diets for Laying Hen’. African Journal of Agricultural Research. Vol. 7(8).
Pp. 1302–1306.
Fletcher, R. 1987. Practical Methods of Optimization. John Wiley and Sons. The Atrium,
Southern Gate, Chichester, ISBN: 9780471915478.
Furuya, T., Satake, T. and Minami, Y. 1997. ‘Evolutionary Programming for Mix
Design’. Computers and Electronics in Agriculture. Vol. 18(3). Pp. 129–135.
Gale, D., Kuhn, H. W. and Tucker, A. W. 1951. ‘Linear Programming and the Theory of
Games’. In Activity Analysis of Production and Allocation (T. C. Koopmans ed.).
New York: Wiley & Sons, Pp. 317–335.
Gallenti, G. 1997. The Use of Computer for the Analysis of Input Demand in Farm
Management: A Multicriteria Approach to the Diet Problem. First European
Conference for Information Technology in Agriculture, 15–18 June 1997.
Garg, M. R., FAO. 2012. ‘Balanced Feeding for Improving Livestock Productivity—
Increase in Milk Production and Nutrient use Effciency and Decrease in Methane
Emission’. FAO Animal Production and Health, Paper No. 173. Rome, Italy.
Garg, M. R., Sherasia, P. L., Bhanderi, B. M. and Makkar, H. P. S. 2016. ‘Nitrogen
Use Effciency for Milk Production on Feeding a Balanced Ration and Predicting
Manure Nitrogen Excretion in Lactating Cows and Buffaloes Under Tropical
Conditions’. Animal Nutrition and Feed Technology. Vol. 16. Pp. 1–12.
Gen, M. and Cheng, R. 1997. Genetic Algorithm and Engineering Design. New York:
John Wiley and Son. ISBN: 0-471-12741-8.
Ghosh, S., Ghosh, J., Pal, D. T. and Gupta, R. 2014. ‘Current Concepts of Feed
Formulation for Livestock using Mathematical Modeling’. Animal Nutrition and
Feed Technology. Vol. 14. Pp. 205–223.
Gill, P. E., Murray, W. and Wright, M. H. 1981. Practical Optimization. London:
Academic Press.
Goldberg, D. E. 1989. Genetic Algorithms in Search, Optimization, and Machine
Learning. Reading, MA: Addison-Wesley.
Goldfarb, D. 1969a. ‘Extension of Davidon’s Variable Metric Method to Maximization
Under Linear Inequality and Equality Constraints’. SIAM Journal on Applied
Mathematics. Vol. 17. Pp. 739–764.
Gosh, S. 2011, September. ‘Least Cost Ration Formulation in Animal Using
Mathematical Modeling: An Integrated Linear and Weighted Goal Programming
Approach’. Unpublished M.Phil. Dissertation. Jain University.
Govt of India. 2012. Ministry of Agriculture Department of Animal Husbandry. New
Delhi: Dairying and fsheries Krishi Bhavan, 19 livestock census all India report.
Guevara, V. R. 2004. ‘Use of Nonlinear Programming to Optimize Performance
Response to Energy Density in Broiler Feed Formulation’. Poultry Science.
Vol. 83(2). Pp. 147–151.
Gupta, R. and Chandan, M. 2013. ‘Use of Controlled Random Search Technique for
Global Optimization” in Animal Diet Problem’. International Journal of Emerging
Technology and Advanced Engineering. Vol. 3(2).
98 References
Gupta, R., Chandan, M. and Kuntal, R. S. 2013a. ‘Heuristic Approaches in Solving

Non-Linear Programming Model of Livestock Ration for Global Optimization’.
International Journal of Engineering Sciences and Emerging Technologies
(IJESET). Vol. 6(1). Pp. 37–48.
Gupta, R., Kuntal, R. S. and Ramesh, K. 2013b. ‘Heuristic Approach to Goal
Programming Problem for Animal Ration Formulation’. International Journal of
Engineering and Innovative Technology (IJEIT). Vol. 3(4). Pp. 414–422.
Haar, V. and M. J. Black. 1991. ‘Ration Formulation Using Linear Programming’.
Veterinary Clinics of North America: Food Animal Practice. Vol. 7. Pp. 541–556.
Hadrich, J. C., Wolf, C. A., Black, J. R. and Harsh, S. B. 2008. ‘Incorporating
Environmentally Compliant Manure Nutrient Disposal Costs into Least-Cost
Livestock Ration Formulation’. Journal of Agricultural and Applied Economics.
Vol. 40(1). Pp. 287–300.
Han, S. P. 1977. ‘A Globally Convergent Method for Nonlinear Programming’. Journal
of Optimization Theory and Applications. Vol. 22. P. 297.
Herrera, F., Lozano, M. and Verdegay, J. L. 1995. ‘Fuzzy Connective based Crossover
Operators to Model Genetic Algorithms Population Diversity. Technical Report
No. DECSAI-95110, University of Granada, 18071 Granada, Spain.
Hertzler, G. 1987. ‘Dynamically Optimal and Approximately Optimal Beef Cattle
Diets Formulated by Nonlinear Programming’. Western Journal of Agricultural
Economics. Vol. 13. Pp. 7–17.
Hertzler, G. 1988. ‘Dynamically Optimal and Approximately Optimal Beef Cattle
Diets Formulated by Nonlinear Programming’. Western Journal of Agricultural
Economics. Vol. 13(1). Pp. 7–17.
Himmelblau, D. M. 1972. Applied Nonlinear Programming. New York: McGraw-Hill.
Hock, W. and Schittkowski, K. 1983. ‘A Comparative Performance Evaluation of 27
Nonlinear Programming Codes’. Computing. Vol. 30. Pp. 335–358.
Holland, J. H. 1975. Adaptation in Natural and Artifcial Systems. Ann Arbor, MI:
University of Michigan Press.
Jean dit Bailleul, P., Rivest, J., Dubeau, F. and Pomar, C. 2001. ‘Reducing Nitrogen
Excretion in Pigs by Modifying the Traditional Least-cost Formulation Algorithm’.
Livestock Production Science. Vol. 72. Pp. 199–211.
Koda, M. 2012. ‘Chaos Search in Fourier Amplitude Sensitivity Test Version’. Journal
of Information and Communication Technology. Vol. 11. Pp. 1–16.
Kumar, J. R. K. 2012. ‘Effect of Polygamy with Selection in Genetic Algorithm’. International
Journal of Soft Computing and Engineering. Vol. 2(1). ISSN: 2231-2307.
Kuntal R.S., Gupta R., Rajendran D., Patil V. 2019. ‘Study of Real-Coded Hybrid Genetic
Algorithm (RGA) to Find Least-Cost Ration for Non-Pregnant Dairy Buffaloes’. In:
J. C. Bansal, K. N. Das, A. Nagar, K. Deep, A. K. Ojha (Eds.), Intelligent Systems and
Computing 817: Soft Computing for Problem Solving, Springer Nature Singapore
Pte Ltd., Pp.369–389. ISBN: 9789811315947.
Kuntal, R. S., Gupta, R., Rajendran, D. and Patil, V. 2016. ‘Application of Real
Coded Genetic Algorithm (RGA) to Find Least Cost Feedstuffs for Dairy Cattle
During Pregnancy’. Asian Journal of Animal and Veterinary Advances. Vol. 11.
Pp. 594–607.
Kuntal, R. S., Gupta, R., Rajendran, D. and Patil, V. 2018. ‘A Goal Programming
Approach to Ration Formulation Problem for Indian Dairy Cows’. International
Journal of Current Advanced Research. Vol. 7(4c). Pp. 11506–11510.
References 99
Lara, P. 1993. ‘Multiple Objective Fractional Programming and Livestock Ration

Formulation: A Case Study for Dairy Cow Diets in Spain’. Agricultural Systems.
Vol. 41. Pp. 321–334.
Leng, R. A. 1991. ‘Feeding Strategies for Improving Milk Production of Dairy Animals
Managed by Small Holder Farmers in the Tropics,’ L. Speedy, A., Sansoucy, R.
(Eds.), Feeding Dairy Cows in the Tropics, FAO, Rome. Pp. 82–104.
Lucasius, C. B. and Kateman, G. 1989. ‘Applications of Genetic Algorithms in
Chemometrics’. In Proceedings ofthe 3rd International Conference on Genetic
Algorithms. Morgan Kaufmann, Los Altos, CA. Pp. 170–176.
MacDonald, Z. 1995. ‘Teaching Linear Programming using Microsoft Excel Solver’.
Computers in Higher Education Economics Review. Vol. 9(3). Pp. 7–10.
McCormick, G. P. 1970. ‘The Variable Reduction Method for Nonlinear Programming’.
Management Science Theory. Vol. 17. Pp. 146–160.
McCormick, G. P. 1983. Nonlinear Programming: Theory, Algorithms, and Applications.
New York: Wiley, P. 444.
Meyer, J. H. and W. N. Garrett. 1967. ‘Effciency of Feed Utilization’. Journal of Animal
Science. Vol. 26. P. 638.
Meyer, K. 1985. ‘Genetic Parameters of Dairy Production of Australian Black and White
Cows’. Livestock Production Science. Vol. 12. P. 205.
Mitani, K. and Nakayama, H. 1997. ‘A Multi-objective Diet Planning Support System
Using the Satisfying Trade-off Method’. Journal of Multi-Criteria Decision
Analysis. Vol. 6. Pp. 131–139.
Mohan, C. and Shanker, K. 1994. ‘A Random Search Technique for Global optimiza-
tion Based on Quadratic Approximation’. Asia Pacifc Journal of Operations
Research. Vol. 11. Pp. 93–101.
Mondal, M. S., Yamaguchi, H., Date, Y., Shimbara, T., Toshinai, K., Shimomura,
Y., Mori, M. and Nakazato, M. 2003. ‘A Role for Neuropeptide W in the Regulation
of Feeding Behavior’. Endocrinology. Vol. 144(11,1). Pp. 4729–4733. https://doi.
org/10.1210/en.2003-0536
Mudgal, V., Mehta, M. K., Rane, A. S. and Nanavati, S. 2003. ‘A Survey on Feeding
Practices and Nutritional Status of Dairy Animals in Madhya Pradesh’. Indian
Journal of Animal Nutrition. Vol. 20(2). Pp. 217–220.
Munford, A. G. 1989. ‘A Microcomputer System for Formulating Animal Diets Which
May Involve Liquid Raw Materials’. European Journal of Operational Research.
Vol. 41. Pp. 270–276.
Munford, A. G. 1996. ‘The Use of Iterative Linear Programming in Practical Applications
of Animal Diet Formulation’. Mathematics and Computers in Simulation. Vol. 42.
Pp. 255–261.
Munford, A. G. 2005. The Ultramix-Professional Feed Formulation and Livestock
Modelling System. Udine, Italy: AGM Systems Ltd.
Nabasirye, M., Mugisha, J. V. T., Tibayungwa, F. and Kyarisiima, C. C. 2011.
‘Optimization of Input in Animal Production: A Linear Programming Approach to
the Ration Formulation Problem’. International Research Journal of Agricultural
Science and Soil Science (ISSN: 2251–0044). Vol. 1(7). Pp. 22:1–226.
National Research Council. 1981. Effect of Environment on Nutrient Requirements of
Domestic Animals. Washington, DC: The National Academies Press.
National Research Council. 2001. Nutrient Requirements of Dairy Cattle (7th rev. ed.).
Washington, DC: National Academic Science.
100 References
Nocedal, J. and Wright, S. J. 2006. Numerical Optimization. Second edition. Springer

Series in Operations Research. Springer Verlag. Springer Science+Business Media,
LLC. ISBN-10: 0-387-30303-0.
Nott, H. and Combs, G. F. 1967. ‘Data Processing Feed Ingredient Composition Data’.
Feedstuffs. Vol. 39. Pp. 21–22.
Nutrient Requirements of Animals-Cattle and Buffalo (ICAR-NIANP). 2013. ISBN:
978-81-7164-139-9.
Oladokun, V. O. and Johnson, A. 2012. ‘Feed Formulation in Nigerian Poultry Farms:
A Mathematical Programming Approach’. American Journal of Scientifc and
Industrial Research. Vol. 3. Pp. 14–20.
Olorunfemi, T. O. S., Aderibigbe, F. M., Falaki, S. O., Adebayo, O. T. and Fasakin, E. A.
2001. ‘An Overview of Linear Programming Application to Least-cost Ration
Formulation in Aquaculture’. Journal of Technology and Science. Vol. 5. Pp. 84–92.
Olson, D. L. and Swenseth, S. R. 1987. ‘A Linear Approximation for Chance-
Constrained Programming’. Journal of the Operational Research Society. Vol. 38(3).
Pp. 261–267.
Ono, I., Kita, H. and Kobayashi, S. 2003. ‘A Real-coded Genetic Algorithm using
the Unimodal Normal Distribution Crossover’. In Advances in Evolutionary
Computing (A. Ghosh and S. Tsutsui, eds.). Natural Computing Series. Berlin,
Heidelberg: Springer. https://doi.org/10.1007/978-3-642-18965-4_8
Panne, C. V. D. and Popp, W. 1963. ‘Minimum-cost Cattle Feed Under Probabilistic
Protein Constraints’. Management Science. Vol. 9. Pp. 405–430.
Patrick, H. and Schaible, P. J. 1980. Poultry: Feed and Nutrition (2nd Edn.), AVI
Publishing, USA. Pp: 417–458.
Pesti, G. M. and Siela, A. F. 1999. ‘The Use of an Electronic Spreadsheet to Solve Linear
and Non-linear “Stochastic” Feed Formulation Problems’. Journal of Applied
Poultry Research. Vol. 8. Pp. 110–121.
Pond, W. G., Church, D. C. and Pond, K. R. 1995. Basic Animal Nutrition and Feeding.
Canada: John Wiley and Sons, Inc.
Powell, M. J. D. 1977. ‘Quadratic Termination Properties of Davidon’s New Variable
Metric Algorithm’. Mathematical Programming. Vol. 12. Pp. 141–147.
Powell, M. J. D. 1978a. ‘The Convergence of Variable Metric Methods for Nonlinearly
Constrained Optimization Calculations’. In Nonlinear Programming 3, (O. L.
Mangasarian, R. R. Meyer and S. M. Robinson, eds.). Academic Press, New York.
Powell, M. J. D. 1978b. ‘A Fast Algorithm for Nonlinearly Constrained Optimization
Calculations’. In Numerical Analysis. Vol. 630 (G. A. Watson, ed.). Lecture Notes
in Mathematics. Springer Verlag, Berlin.
Pratiksha, S. 2011. Optimization Techniques for Animal Diet Formulation, Gate2Biotech,
e-portal, February 2011, Vol. 1(2). Pp. 1–5, ISSN 1802-2685, Europe.
Radhika, V. and Rao, S. B. N. 2010. ‘Formulation of Low-cost Balanced Ration for
Livestock Using Microsoft Excel’. Wayamba Journal of Animal Science.
Pp. 38–41.
Ranjan, S. K. 1998. Nutrient Requirements of Livestock and Poultry. New Delhi: Indian
Council of Agricultural Research.
Reddy, A. A., Rani, C. R., Cadman, T., Kumar, S. N. and Reddy, A. N. 2016. ‘Towards
Sustainable Indicators of Food and Nutritional Outcomes in India. World Journal of
Science, Technology and Sustainable Development. Vol. 13(2). Pp. 128–142. https://
doi.org/10.1108/WJSTSD-10-2015-0049
References 101
Rehman, A. R. 2014. ‘Evolutionary Algorithms with Average Crossover and Power

Heuristics for Aquaculture Diet Formulation’. Ph.D. Thesis. Sintok, Malaysia:
University Utara Malaysia.
Rehman, T. and Romero, C. 1984. ‘Multiple-criteria Decision-making Techniques and Their
Role in Livestock Ration Formulation’. Agricultural Systems. Vol. 15. Pp. 23–49.
Rehman, T. and Romero, C. 1987. ‘Goal Programming with Penalty Functions and
Livestock Ration Formulation’. Agricultural Systems. Vol. 23. Pp. 117–132.
Romero, C. and Rehman, T. 1984. Goal Programming and Multiple Criteria Decision-
making in Farm Planning: An Expository Analysis. Journal of Agricultural
Economics. Vol. 35. Pp. 177–190. doi:10.1111/J.1477-9552.1984.TB02045.X.
Rose, S. P. 1997. Principles of Poultry Science. CAB International, Wallingford, Oxon,
ISBN: 0 85199 122X.
Rossi, R. 2004. ‘Least-cost Formulation Software: An Introduction’. Aqua Feed Formul
Beyond. Vol. 3. Pp. 3–5.
Roush, W. B., Stock, R. H., Cravener, T. L. and D'Alfonso, T. H. 1994. ‘Using Chance-
Constrained Programming for Animal Feed Formulation at Agway’. Interfaces.
Vol. 24(2). Pp. 53–58.
Şahman, M. A., Çunkaş, M., İnal, Ş., İnal, F., Coşkun, B. and Taşkiran, U. 2009. ‘Cost
Optimization of Feed Mixes by Genetic Algorithms’. Advances in Engineering
Software. Vol. 40(10). Pp. 965–974.
Saxena, P. 2011a. ‘Comparison of Linear and Non-linear Programming Techniques
for Animal Diet’. Applied Mathematics. Vol. 1(2). Pp. 106–108. https://doi.
org/10.5923/j.am.20110102.17
Saxena, P. 2011b. ‘Application of Nonlinear Programming for Optimization of Nutrient
Requirements for Maximum Weight Gain in Buffaloes’. International Journal of
Food Science and Nutrition Engineering. Vol. 1(1). Pp. 8–10.
Saxena, P. 2011c. ‘Optimization Techniques for Animal Diet Formulation’. www.
Gate2Biotech.com. Vol. 1(2). Pp. 1–5.
Sebastian, C., Akinnifesi, F. K., Ajayi, O., Sileshi, G., Mngomba, S. and Gondwe, F.
M. T. 2008. ‘A Simple Method of Formulating Least-cost Diets for Smallholder
Dairy Production in Sub-Saharan Africa’. African Journal of Biotechnology. Vol. 7.
Pp. 2925–2933.
Shilpa, J., Dinesh, B. and Prakash, C. 2013. ‘Comparative Analysis of Real and Binary
Coded Genetic Algorithm for Fuzzy Time Series Prediction’. International
Journal of Education and Information Sciences, Vol. 3. Pp. 299–304.
Singh, K. P. and Singh, I. 2015. ‘Buffalo Diversity in India: Breeds and Defned
Populations’. Dairy Year Book. (2014–15) Pp. 33–36.
Sklan, D. and Dariel, I. 1993. ‘Diet Planning for Humans Using Mixed-integer Linear
Programming’. British Journal of Nutrition. Vol. 70. Pp. 27–35.
Taha, H.A. 1987. Operations Research (4th Edn.). United States edition: Macmillan
Publishing Co. P. 876.
Tedeschi, L. O., Fox, D. G., Chase, L. E. and Wang, S. J. 2004. ‘Whole-herd Optimization
with the CornellNet Carbohydrate and Protein System: Predicting Feed Biological
Values for Diet Optimization Withlinear Programming’. Poultry Science. Vol. 83.
Pp. 147–151.
Tedeschi, L. O., Fox, D. G., Sainz, R. D., Barioni, L. G., de Medeiros, S.
R. and Boin, C. 2005. Mathematical Models in Ruminant Nutrition. Scientia
Agricola. Vol. 62. Pp. 76–91.
102 References
Townsley, R. 1968. ‘Derivation of Optimal Livestock Rations Using Quadratic

Programming. Journal of Agricultural Economics. Vol. 19(3). Pp. 347–354.
Tozer, P. R. 2000. ‘Least-Cost Ration Formulations for Holstein Dairy Heifers By
Using Linear and Stochastic Programming’. Journal of Dairy Science. Vol. 83.
Pp. 443–451.
Tozer, P. R. and Stokes, J. R. 2001. ‘A Multiobjective Programming Approach to Feed
Ration Balancing and Nutrient Management’. Agricultural Systems. Vol. 67(3).
Pp. 201–215.
Van de Panne, C. and Popp, W. 1963. ‘Minimum-Cost Cattle Feed Und- er Probabilistic
Protein Constarints’. Management Science. Vol. 9(3). Pp. 405–430.
Xiong, B-H., Luo, Q-Y. and Pang, Z-H. 2003. Application of Dual Model in Animal
Feed Formulation optimization System. Scientia Agricultura Sinica. Vol. 2003.
Pp. 1347–1351.
Waugh, F. 1951. ‘The Minimum-Cost Dairy Feed’. Journal of Farm Economics. Vol. 33.
Pp. 299–310.
Zgajnar, J., Juvancic, L. and Kavcic, S. 2009. ‘Combination of Linear and Weighted
Goal Programming with Penalty Function in Optimization of Daily Dairy Cow
Ration’. Agricultural Economics—Czech. Pp. 492–500.
Zgajnar, J. and Kavcic, S. 2008. ‘Spreadsheet Tool for Least-cost and Nutrition Balanced
Beef Ration Formulation’. Acta Agriculturae Slovenia. Vol. 2(Suppl.). Pp. 187–194.
Zgajnar, J. and Kavcic, S. 2009. ‘Multi-goal Pig Ration Formulation; Mathematical
Optimization Approach’. Agronomy Research. Vol. 7. Pp. 775–782.
Zhang, F. and Roush, W. B. 2002. ‘Multiple-objective (Goal) Programming Model for
Feed Formulation: An Example for Reducing Nutrient Variation’. Poultry Science.
Vol. 81. Pp. 182–192.
Zheng, D. W., et al. 1996. ‘Evolution Program for Nonlinear Goal Programming’. 18th
International Conference on Computers and Industrial Engineering.
Index
Note: Numbers in bold indicate tables and those in italics indicate fgures.
A Basic Animal Husbandry & Fisheries

Statistics 2017, 2, 61
Additivity, 11 Berseem
Adugodi, Bangalore 5, 26 least cost and deviation value, hybrid RGA,
Animal feed formulation, see Feed GP Model-1, 82
formulation least cost and deviation value, hybrid RGA,
Animal nutrition, 1–6 GP Model-2, 83
“As fresh basis” least cost and deviation value, hybrid RGA,
Least Cost of Ration Cattle-1 (LP Model), 54 GP Model-3, 84
Least Cost of Ration Cattle-2 (LP Model), 56 least cost ration, dairy buffalo, ‘as fresh’
Least Cost of Ration Cattle-3 (LP Model), 57 basis, 81
Least Cost Ration for Non-Pregnant Dairy least cost ration, dairy buffalo, dry matter
Buffalo (LP Model), 81 (DM) basis, 79
Net Proft Per Cattle for 10 Litres of Milk nutrient content, 6
Each Day, 58 nutrient content on dry matter (DM) basis, 73
real coded hybrid genetic algorithm BGA see Binary coded genetic algorithm
(RHGA), results and discussion, 79, Binary coded genetic algorithm (BGA), 52,
80–81, 90 87, 88
total mixed ration (TMR), 81, 84, 85 comparative analysis of real (RGA) and, 62,
65–67
B limitations of, 88
ration formulation problem solved by, 27–43
Bajra, 5 Bran Max (b), 35
Bajra Napier Co-4 (“Co-4”) grass, 47 Broken rice, 5
least cost and deviation value, hybrid RGA,
GP Model-1, 82
least cost and deviation value, hybrid RGA, C
GP Model-2, 83 Calcite, 47
least cost and deviation value, hybrid RGA, least cost and deviation value, hybrid RGA,
GP Model-3, 84 GP Model-1, 82
least cost, deviation, and constraints values, least cost and deviation value, hybrid RGA,
hybrid RGA, 66 GP Model-2, 83
least cost ration, dairy buffalo, ‘as fresh’ least cost and deviation value, hybrid RGA,
basis, 81 GP Model-3, 85
least cost ration, dairy buffalo, dry matter least cost, deviation, and constraints values,
(DM) basis, 79 hybrid RGA, 66
nutrient content, 6 least cost ration, dairy buffalo, ‘as fresh’
nutrient content on dry matter (DM) basis, basis, 81
48, 73 least cost ration, dairy buffalo, dry matter
Bangalore (DM) basis, 80
Adugodi, 5, 26 nutrient content, 6
National Institute of Animal Nutrition and nutrient content on dry matter (DM) basis,
Physiology, 72, 74, 86 48, 73
103
104 Index
Cereal grasses, 5 least cost and deviation value, hybrid RGA,

Certainty, 11 GP Model-3, 85
Chickpea husk least cost, deviation, and constraints values,
least cost and deviation value, hybrid RGA, hybrid RGA, 66
GP Model-1, 82 least cost ration, dairy buffalo, ‘as fresh’
least cost and deviation value, hybrid RGA, basis, 81
GP Model-2, 83 least cost ration, dairy buffalo, dry matter
least cost and deviation value, hybrid RGA, (DM) basis, 80
GP Model-3, 85 nutrient content on DM basis, 48, 73
least cost, deviation, and constraints values, Cotton seed, 47
hybrid RGA, 66 least cost and deviation value, hybrid RGA,
least cost ration, dairy buffalo, ‘as fresh’ GP Model-1, 82
basis, 81 least cost and deviation value, hybrid RGA,
least cost ration, dairy buffalo, dry matter GP Model-2, 83
(DM) basis, 80 least cost and deviation value, hybrid RGA,
nutrient content, 6 GP Model-3, 85
nutrient content on DM basis, 48, 73 least cost, deviation, and constraints values,
Co-FS 29 sorghum, see Sorghum hybrid RGA, 66
Composition of Concentrate Mixtures least cost ration, dairy buffalo, ‘as fresh’
in Respect of Dicalcium Phosphate basis, 81
(DCP) and Total Digestible Nutrients least cost ration, dairy buffalo, dry matter
(TDN), 28 (DM) basis 80
Comparison of Solution Obtained by GA and nutrient content, 6
RST for NLP Model, 31 nutrient content on DM basis, 48, 73
Comparison of Decision Variables by RST Cotton seed cake, 28, 28
and GA 38 Cowpea, 34
Bharti’s update of Price, 16 CP, see Crude protein
comparison of decision variables using GA Cross-breed cattle, 3
and, 38 Cross-breed cows, 3
comparison of GA and, 41 Cross-breeding programs, 27
maximum milk yield using GA and, 32, 33 Crossover
Price’s presentation of, 16 1-point, 25
RST2, 29, 30, 31, 42, 88 2-point, 25
Copra de-oiled cake (DOC), 47 evolutionary algorithms (EA) and, 23
least cost and deviation value, hybrid RGA, Furuya’s use of new form of, 71
GP Model-1, 82 genetic algorithms and, 25, 52
least cost and deviation value, hybrid RGA, heuristic, 59, 78, 90
GP Model-2, 83 mutation and, 20, 25, 65, 71, 93
least cost and deviation value, hybrid RGA, real coded genetic algorithm (RGA) with/
GP Model-3, 84 without, 79, 80, 81, 86, 93
least cost, deviation, and constraints values, real coded hybrid genetic algorithm (RGHA)
hybrid RGA, 66 and, 78
least cost ration, dairy buffalo, dry matter Crossover children, 53
(DM) basis 80 “CrossoverFcn” options, 25
least cost ration, dairy buffalo, ‘as fresh’ “Crossoverheuristic,” 25
basis, 81 Crossover method, heuristic, 53
nutrient content, 6 Crossover offspring, creation of, 25, 30, 52,
nutrient content on DM basis, 48, 73 53, 65, 77, 78
Cotton de-oiled cake (DOC), 47 Crossover operator, 53, 77, 78
least cost and deviation value, hybrid RGA, Crude protein (CP), 4, 28, 45, 47, 48, 58
GP Model-1, 82 GP Model-1, maximizing of, 63
least cost and deviation value, hybrid RGA, GP Model-2, maximizing of, 64
GP Model-2, 83 target values on dry matter basis, 62
Index 105
D Department of Animal Husbandry, Dairying,

and Fisheries, 2, 70
Dairy animals in India, 1, 27 Annual Report, 71
nutrients important for, 3 Dicalcium phosphate (DCP), 47
Dairy buffaloes, non-pregnant, 69–86 least cost and deviation value, hybrid RGA,
Dairy cattle/cows GP Model-1, 82
feed for, 5 least cost and deviation value, hybrid RGA,
feeding of, 4 GP Model-2, 83
formulating rations in least cost manner for, 20 least cost and deviation value, hybrid RGA,
improvement of milk yield of, 42 GP Model-3, 85
nutrient excretion by, 19 least cost, deviation, and constraints values,
nutrient requirements set by, 26 hybrid RGA, 66
shortage of feed ingredients for, 27 least cost ration, dairy buffalo, ‘as fresh’ basis, 81
Dairy cattle/cows during pregnancy, least-cost least cost ration, dairy buffalo, dry matter
feed formulation for 45–68 (DM) basis, 80
Dairy farms nutrient content, 6
commercial, 9 nutrient content on DM basis, 48, 73
proftability of, 12 Divisibility, 11
Dairy farmers DM, see dry matter
Karnataka, 7 DMI, see dry matter intake
Dairy farming, 2 dry/green roughage, 34, 35
Dairy feed Dry matter (DM), 4, 6, 81, 87, 89
least-cost, formulation of 14, 46 concentrated, 16
Dairy products, 3 forage, 16
Darwin, [Charles], 30 total mixed ration (TMR) cost on, 90
DCP, see Dicalcium phosphate Dry matter intake (DMI), 46, 47, 70, 73
Decision analysis, see Multi-criteria decision Dry roughage, 5
analysis (MCDA)
Decision maker, 13, 14, 62, 91–92
Decision-making, 1 E
fexibility in, 72 EA, see Evolutionary algorithm
linear programming applied to, 10 Elitism
Decisions convergence, importance to, 25
dynamically optimal, 17 LP model solved with adaptability
feeding and selling, 17 feasibility mutation, heuristic crossover,
judicial, 72 and, 59, 80, 90
Decision variables, 42 Elitism operators, 78
binary coded GA representing, 65 Energy, 4
comparison by RST and GA, 38 Evolutionary algorithm (EA), 53, 59, 87
GRG nonlinear algorithm dealing with constituent parts of, 61, 75
problems involving, 58, 89 as heuristic, 89
linear programming (LP) model and, linear programming (LP) model solved
11–13 using, 46, 58, 89
non-negative, 22 linear programming problem (LPP) solved
real coded hybrid genetic algorithm using, 74
involving, 77–78 in MS-Solver, 21, 22–23, 58
variations in values of, 88 Evolutionary theory, 65
Deviations for Linear Model of Level-1, Excel, see MS Excel Solver
Level-2 and Level-3, 39
Deviations for Nonlinear GP Model of
F
Level-1, Level-2 and Level-3, 40
Deviation variables, 18, 33–34, 41, 61, Feed and feeding costs, high
63, 91 impacts of, 17, 61, 70–72
106 Index
Feed formulation Finiteness, 11

broiler, 17 “Fmincon” command, 65, 69, 78, 79, 91
defnition of, 9 Fodder, see dry fodder; green fodder; maize
least cost, evolutionary algorithm (EA) not fodder
so accurate for solving 89 Fodder crops, area under 2, 61
least-cost for dairy cattle during pregnancy,
45–68
G
Least Cost Feed Formulation of Dairy Cattle, 49
linear and nonlinear programming model for GA, see genetic algorithm
11, 28–32, 42 General reduced gradient (GRG)
linearity and, 11 LP model solved by, 46, 89
linear programming applied to, 46 nonlinear, 53, 54–58, 59, 60, 62, 87, 90, 92
linear programming (LP) combined with nonlinear in MS-Excel Solver, 22
goal programming (GP) for, 20 Genetic algorithm (GA), 10, 30, 61, 65
linear programming software for, 13 canonical, 20
livestock, 11 classical methods, difference from, 24
multi-objective programming applied to, 46 ‘crossover’ as important parameter of, 25
non-linear model applied to, 42 defned as (inputs for), 24–25
primary objective of, 17 extended work using, 29
software for, 13 feed formulation problem solved using, 52
stochastic programming applied to, 17 Furuya’s use of, 61
real coded genetic algorithm used to solve, heuristic technique of, 5
45–68 idea behind, 30
Feed formulation problem inputs for, 24–26
evolutionary algorithm (EA) to solve, 58 local versus global minima stipulated by, 78
genetic algorithm (GA) to solve, 52 MATLAB and, 23
FEED FORMULATION software, 13 overview and defnition of, 19–20
Feeding costs, see Feed and feeding costs, high and random search technique (RST),
Feeding practices, 70–72 comparison of, 41
appropriate, 72 Sahman’s use of, 61
studies of, 61 ‘schema theorem,’ 20
Feeding standards solution of nonlinear programming model
National Research Council, as set by, 29 by, 30
Feed intake, 16 Solutions of nonlinear programming
FEEDLIVE, 13 problems (NLPP) using, 42
Feed mix, 86, 92 see also, Binary coded genetic algorithm
improving, 68 (BGA); Evolutionary algorithm; Real
Feed mixing problem coded genetic algorithm (RCGA)
linear programming (LP) insuffcient to Genetic operators, 20
solve, 14 Global vs. local minima, 78–79
FEEDMU, 13 Goal programming (GP)
FEEDMU2, 13 goal of, 71
Feed requirements, 91 high priority objectives of, 70
Feedstuffs linear programming (LP) combined with,
‘as fresh basis,’ 59, 81 46, 60
classifcation of, 5 main reason to apply, 61
dry season, decline in quality during, 70 multiple goal programming (MGP), 19
limited availability of, Karnataka, 60, 61 nonlinear, 34
nutrient content of, 47, 48, 73 nonlinear, with priority ranked goals, 35
RHGA to solve, 81–84 overview and discussion of, 18–19
shortages of, 7 real coded hybrid genetic algorithm used to
Fetus growth, 7, 46–47, 87 solve, 62
Index 107
Goal programming (GP) approach healthy, 3

31 decision variables, 65 perennial, 5
ration formulation problem in dairy cows Grazing lands, decrease of 2, 61
and, 60–68 Green fodder, 5, 28
real coded hybrid genetic algorithm used to Green roughage, 5, 6, 47
solve, 62 Grit, 6
seven goals formulated as goal functions, GRG, see generalized reduced gradient
63, 65 Groundnut cake, 28, 28
Zoran-Babic et al., 61
Goal programming (GP) Model, 1, 63, 91–92 H
deviations for nonlinear models of, 40
least cost and deviation solved by hybrid Heuristic algorithms, 2, 27, 29, 30
RGA for, 66, 82 evolutionary algorithm (EA), 89
priority values, 76 Heuristic approach, 29, 42
Goal programming (GP) Model 2, 63, 91–92 Heuristic crossover method, 53, 78
deviations for nonlinear models of, 40 adaptive feasibility mutation and, 59
least cost and deviation solved by hybrid Heuristic techniques
RGA for, 66, 83 genetic algorithm, 5, 77
priority values, 76 real coded genetic algorithm (RGA), 87
Goal programming (GP) Model 3, 64, 91–92
deviations for nonlinear models of, 40 I
least cost and deviation solved by hybrid
RGA for, 66, 84 ICAR see Indian Council of Agricultural
Goal programming problem (GPP) Research
general mathematical formula for, 18 Indian Council of Agricultural Research
linear, 42 (ICAR), 3, 26, 33, 62, 74, 87
nonlinear, 20, 42 Indian Council of Agricultural Research-
GP, see Goal programming National Institute of Animal Nutrition
Gradient-based search technique, 79; see also, and Physiology (ICAR-NIANP), 5, 33,
“Fmincon” 46, 47, 70, 86
Gradients, see General reduced gradient INFORMS, see Institute for Operations
(GRG) Research and the Management Sciences
Gran chunnies, 47 Institute for Operations Research and the
least cost and deviation value, hybrid RGA, Management Sciences (INFORMS), 1
GP Model-1, 82
least cost and deviation value, hybrid RGA, K
GP Model-2, 83
least cost and deviation value, hybrid RGA, Karush-Kuhn-Tucker (KKT) conditions, 58, 89
GP Model-3, 85 Karnataka, dairy farmers in, 2, 7
least cost, deviation, and constraints values, Mandya district, 26, 45–47, 60–61, 69,
hybrid RGA, 66 86, 89
least cost ration, dairy buffalo, ‘as fresh’ Kuhn-Tucker (KT) conditions, 15, 29, 42, 88
basis, 81
least cost ration, dairy buffalo, dry matter L
(DM) basis, 80
nutrient content, 6 Lagrangian function, 79
nutrient content on dry matter (DM) basis, Least Cost and Deviation Value Solved by
48, 73 Hybrid RGA for GP Model-1, 82
Grass; see also, Bajra Napier Co-4 (“Co-4”) Least Cost and Deviation Value Solved by
grass Hybrid RGA for GP Model-2, 83
annual, 5 Least Cost and Deviation Value Solved by
cereal, 5 Hybrid RGA for GP Model-3, 84
108 Index
Least Cost Balanced Ration for Non-Pregnant Mitani and Nakayama on, 14
Dairy Buffalo, 74 overcoming limitations of, 14, 33
Least Cost, Deviation, and Constraints ration formulation and, 10–14
Value Solved by Hybrid RGA for Goal Rehman and Romero on, 61, 71
Programming Models, 66 Right-hand-side (RHS) presumptions
Least-cost feed formulation for dairy cattle/ of, 71
cows during pregnancy, 45–68 simplex method, 11, 29, 33, 62, 89, 92
Least Cost Feed Formulation of Dairy simplex method in MS-Excel, 22
Cattle, 49 Linear Programming Problems (LPP) 60,
Least Cost of Ration by Various Techniques 69; see also, nonlinear programming
for Cattle-1 on As Fresh Basis (LP problems (NLPP)
Model), 54 Level-1, 2 and 3 by genetic algorithm
Least Cost of Ration by Various Techniques (GA), 36
for Cattle-1 on DM Basis, 54 Linear programming (LP) techniques, 46
Least Cost of Ration by Various Techniques Livestock census, All India report, 2, 60, 70
for Cattle-2 on As Fresh Basis (LP
Model), 56 M
Least Cost of Ration by Various Techniques
for Cattle-2 on DM Basis (LP Model), 55 Maize, 5; see also, Maize fodder, Maize
Least Cost of Ration by Various Techniques stover
for Cattle-3 on As Fresh Basis, 57 Maize fodder, 47
Least Cost of Ration by Various Techniques least cost and deviation value, hybrid RGA,
for Cattle-3 on DM Basis (LP Model), 56 GP Model-1, 82
Least Cost Ration Formulated by Various least cost and deviation value, hybrid RGA,
Techniques for Non-Pregnant Dairy Buffalo GP Model-2, 83
on “As Fresh Basis” (LP Model), 81 least cost and deviation value, hybrid RGA,
Least Cost Ration Formulated by Various GP Model-3, 84
Techniques for Non-Pregnant Dairy least cost, deviation, and constraints values,
Buffalo on DM Basis (LP Model), 79 hybrid RGA, 66
Legumes, 5, 34, 35 least cost ration, dairy buffalo, ‘as fresh’
Linear interactive and discrete optimizer basis, 81
(LINDO), 13 least cost ration, dairy buffalo, dry matter
Linearity, 11 (DM) basis, 79
Linear programming based software, 13 nutrient content, 6
Linear programming (LP) model, 87 nutrient content on dry matter (DM) basis,
animal feed formulation, history of use in, 12 48, 73
assumptions informing, 11–12 Maize stover, see Maize fodder
cattle-1, 50, 90, 91 Markov process, 2
cattle-2, 50, 90, 91 MATLAB (Matrix-Laboratory), 13, 21,
cattle-3, 51, 90, 91 23–24, 25, 52
computation of, using MS-Excel Solver, ‘fmincon,’ 79
22–25 ‘gaoptimset,’ 65, 77
dairy buffalo, 74, 79, 80, 81 Maximum Milk Yield Using GA (Original
Danzig, George, 10–11 Bounds), 32
goal programming (GP) combined with, Maximum Milk Yield Using GA (Reduced
20, 46 Bounds), 32
goal programming model and, 68, 72 Maximum Milk Yield Using RST and GA
history of, 10–11 (Original Bounds), 32
hybrid genetic analysis (GA) used to Maximum Milk Yield Using RST and GA
solve, 92 (Reduced Bounds), 33
Kuntal’s improvements on model, 68 MCDA, see Multi-criteria decision analysis
limitations in ration formulation, 13–14, MCDM process, see Multi-criteria decision
19, 61 making (MCDM) process
Index 109
Milk (milch) linear programming problems (LPP) solved

cost per litre, 69 using, 46
Karnataka’s per capita availability of, 61 optimum metabolizable energy found using, 17
Milk fat, 4, 47, 69 sequential quadratic programming (SQP)
4%, 60, 62, 87, 89 for, 79
6%, 70, 72, 73, 87, 91 Multi-criteria decision analysis (MCDA), 18
buffalo, 71, 72, 73 Multi-criteria decision making (MCDM)
Milk production process, 18
cattle and buffalo as primary animals Multi-criteria decision-making technique, 19
for, 27 Multi-objective goal programming, 14
effcient and proftable model of, 12 Multi-objective programming, 12, 20
feed costs as largest expense in, 9 Tozer’s use of, 46
Karnataka’s share of, 61 Zhang’s use of, 46
Indian economy and, 7 Mutation, see Crossover
nutrients required for, 10
overfeeding of animals and, 7 N
Milk production rates
cows and buffalo, 2–3 National Dairy Research Institute (NDRI), 28
Milk rate in dairy 58 National Institute of Animal Nutrition and
Milk yield in cows Physiology (NIANP), 5; see also, Indian
maximizing, 28 Council of Agricultural Research-
maximum, using GA, 32 National Institute of Animal Nutrition
maximum, using RST and GA, 33 and Physiology (ICAR-NIANP)
variables impacting, 28–29 National Research Council (NRC) standards
weight gain and, 18 calcium and phosphorus, 4–5
Milk yield in pregnant cows feeding standards established by, 29
linear programming model (LP) for three nutrient requirement for buffalo set by, 74,
different types of, 46 86, 86
Mineral mixture (MM) nutrient requirements for dairy cattle set by,
concentrate mixtures, 28 26, 46, 47, 62
least cost and deviation value, hybrid RGA, NDRI, see National Dairy Research Institute
GP Model-1, 82 Net Proft Per Cattle in Indian rupees (INR)
least cost and deviation value, hybrid RGA, to Farmers for 10 Litres of Milk on As
GP Model-2, 83 Fresh Basis Each Day, 58
least cost and deviation value, hybrid RGA, Newton and Quasi-Newton-type methods,
GP Model-3, 85 15, 79
least cost, deviation, and constraints values, NIANP, see National Institute of Animal
hybrid RGA, 66 Nutrition and Physiology
least cost ration, dairy buffalo, ‘as fresh’ NLPP, see Nonlinear Programming Problem
basis, 81 Nonlinear programming problems (NLPP),
least cost ration, dairy buffalo, dry matter 15, 17
(DM) basis, 80 animal diet formulation, 18
nutrient content, 6 constrained, 16
nutrient content on dry matter (DM) basis, evolutionary algorithms and, 22
48, 73 solution set for Level-1, 2 and 3 by genetic
Minerals, 4–5 algorithm (GA), 37–38, 42, 88
Minima, see Global vs. local minima unconstrained, 16
MS (Microsoft) Excel Solver, 13 Nonlinear GP Model with Priority Ranked
computation of linear programming (LP) Goals for Animal at Level-1 to 3 with
model using, 21–23, 58, 89 Objective Function (z) to Different Goals
evolutionary algorithm in, 22–23 Which Need to Be Minimized, 35
“fmincon” of 69, 79 Nonlinear Programming (NLP) Model by GA
GRG nonlinear in, 22 for Original Bounds (Test Problem-1), 30
110 Index
Nonlinear Programming (NLP) Model by GA O

for Reduced Bounds (Test Problem-2), 31
Nonlinear programming problems (NLPPs), Objective function, 11
15, 17, 42, 88 as convex functions, 15
animal diet formulation, NLPPS used to goal programming and, 18
solve, 18 inputs for genetic algorithm (GA), 24
evolutionary algorithm used with, 22 linear programming model and, 61
constrained, 16 maximum milk yield using genetic
unconstrained, 16 algorithm (GA), 32, 42
Nonlinear Programming Problem (NLPP) for NLP problem by genetic algorithm (GA) for
Level-1, 2 and 3 by GA, 37 Original Bounds, 30–31
Non-negativity, 11 nonlinear, 29, 34
NRC, see National Research Council real coded hybrid genetic algorithm and,
Nutrient Content of Common Feeds and 77–78
Fodder for Dairy Cattle and Buffalo on single, 27
DM Basis, 6 singularity of, 14
Nutrient Content of Feedstuffs on DM Basis, valley for minimizing, 58, 89
48, 73 Objective function value, 65, 67
Nutrient excretion by dairy cows, 4, 19, 46 Oil seed cake, 5
Nutrient Requirements for Non-Pregnant Offspring, see Crossover offspring
Buffalo Weighing 450 kg and Yielding Operations/Operational research, 1–2
10 Litres of Milk with 6% Fat, 72 Optimization, 12, 87
Nutrient Requirements of Cattle-1, 2, and 3, 47 constrained, 15
Nutrients defnition of, 10
animals’ capacity to digest and mathematical, 1
assimilate, 10 multi-objective, 18, 19
diet formulation and, 6 numerical, 20
costlier, 58, 89 portfolio, 10
essential, 3–4, 9, 70 real space, 52
major, 9–10 unconstrained, 15
milk yields and, 28 Optimization method
overfeeding of, 7 genetic algorithm (GA), 30
ratios of, control over, 16 Optimization model
under-delivery of, 17 nonlinear programming, 17
Nutrients levels, fxed, 14 Optimization problems, 2, 10, 24, 25
Nutrients requirements, calculating, 26 constrained, 23
Nutrients resource, price of, 12 global, 16
Nutrients supplements, 89 linear, 25
Nutrients supplies, imbalance of, 13 nonlinear, 15, 25
Nutrition, see Animal nutrition unconstrained, 23
Nutritional disorder, 4 Optimization techniques, 5
Nutritional requirements Optimization toolbox, 23
changes in diet, impact on, 16
daily, 87 P
ICAR’s publication of, 3
Kearl’s data on, 3, 26 Paddy straw, 6, 47
Lara and Romero’s work on, 19 least cost and deviation value, hybrid RGA,
meeting of, 13, 14 GP Model-1, 82
over-rigid specifcation of, 14 least cost and deviation value, hybrid RGA,
Nutritional values, 7 GP Model-2, 83
Nutritive Values of Indian Cattle Feeds and least cost and deviation value, hybrid RGA,
the Feeding of Animals” (Sen), 3 GP Model-3, 84
Index 111
least cost, deviation, and constraints values, Indian buffaloes in Karnataka, RHGA to
hybrid RGA, 66 solve, 86
least cost ration, dairy buffalo, ‘as fresh’ linear programming in, 10–14
basis, 81 RHGA developed to solve, 86, 91
least cost ration, dairy buffalo, dry matter Real coded genetic algorithm (RGA), 87
(DM) basis, 79 as heuristic technique, 58, 59, 87, 90
nutrient content, 6 hybrid, 70, 82, 83, 84
nutrient content on dry matter (DM) basis, with hybrid function, 65–67, 68, 79, 80, 81,
48, 73 86, 92
Priority Values for GP Model-1, 76 least cost feed formulation for dairy cattle
Priority Values for GP Model-2, 76 using, 47–68
Plots, see Representation of plots least cost ration for non-pregnant dairy
Pregnancy of cows and buffalo, see also, fetus buffaloes, 69–87
growth linear programming (LP) model solved by,
energy requirements, 5, 10 46, 89
least cost feed formulation using RGA numerical optimization using, 52
45–68, 87 offspring provided by, three types of, 53
required nutrients as fxed constraint, 5 outline of steps, 52–53
Proportionality, 11 with random number generation (RNG), 54,
Protein, 3; see also, Crude protein (CP) 55, 56, 57, 58
with seeding, 46, 59, 87, 89, 90
Q without seeding, 54, 55, 56, 57, 58, 59, 87
Real coded hybrid genetic algorithm
Queuing theory, 2 (RHGA), 60
development of, 77–78
R effciency and effectiveness of, 86, 91
least cost ration for buffaloes, used to
Ragi straw, 6, 47 formulate, 69–86
least cost and deviation value, hybrid RGA, limits of binary coded GA overcome by, 87
GP Model-1, 82 ration formulation problem solved by, 86, 91
least cost and deviation value, hybrid RGA, total mixed ration, used to cut down, 70
GP Model-2, 83 Representation of plots, 77–78
least cost and deviation value, hybrid RGA, RGA, see Real coded genetic algorithm
GP Model-3, 84 RHGA, see Real coded hybrid genetic
least cost, deviation, and constraints values, algorithm
hybrid RGA, 66 RNG, see Random number generation
least cost ration, dairy buffalo, ‘as fresh’ RST, see Controlled random search technique
basis, 81 Roughage, defnition of 5; see also, dry
least cost ration, dairy buffalo, dry matter roughage; green roughage
(DM) basis, 79
nutrient content, 6 S
nutrient content on dry matter (DM) basis,
48, 73 Salt, 47
Random number generation concentrate mixtures, 28
real coded genetic algorithm (RGA) with, least cost and deviation value, hybrid RGA,
54, 55, 56, 57, 58 GP Model-1, 82
Ration formulation problem least cost and deviation value, hybrid RGA,
advantages of using spreadsheets to GP Model-2, 83
solve, 21 least cost and deviation value, hybrid RGA,
binary coded genetic algorithm to solve, GP Model-3, 85
27–43 least cost, deviation, and constraints values,
goal-programming model to solve, 33–43 hybrid RGA, 66
112 Index
least cost ration, dairy buffalo, ‘as fresh’ basis, 81 least cost and deviation value, hybrid RGA,
least cost ration, dairy buffalo, dry matter GP Model-3, 84
(DM) basis, 80 least cost, deviation, and constraints values,
nutrient content, 6 hybrid RGA, 66
nutrient content on DM basis, 48, 73 least cost ration, dairy buffalo, ‘as fresh’ basis, 81
Schematic Diagram Representing least cost ration, dairy buffalo, dry matter
Methodology Followed for Least Cost (DM) basis, 79
Feed Formulation of Dairy Cattle, 49 nutrient content, 6
Selected feedstuffs 47; see also, feedstuffs nutrient content on dry matter (DM) basis,
sequential quadratic programming (SQP), 79 48, 73
Simple objective, 11 Stochastic programming
Simplex linear programming (LP), 53, 62 defnition of, 17
MS-Excel Solver, 22 feed formulation method using, 17
Simplex method, 11 linear, 12
controlled random search technique using, 16 WINFEED, 13
FEEDMU, 13 Stochastic techniques, 16
least cost of rations, results per 54–57 Survival of the fttest, principle of, 30
linear programming (LP) problems solved
by, 15, 29, 33, 46, 60 T
Soda bicarbonate, 6
Solution for NLP Model by GA for Original TDN see Total digestible nutrients
Bounds (Test Problem-1), 30 Total digestible nutrients (TDN), 4, 87, 89, 90
Solution for NLP Model by GA for Reduced in common feeds and fodder, 6
Bounds (Test Problem-2), 31 concentrates high in, 5
Solution Set of LPP for Level-1, 2 and 3 by evolutionary algorithm (EA) method, 58
GA, 36 least cost and deviation value, hybrid RGA,
Solution Set of the Nonlinear Programming GP Model-1, 83
Problem (NLPP) for Level-1, 2 and 3 by least cost and deviation value, hybrid RGA,
GA, 37 GP Model-2, 84
Sorghum least cost and deviation value, hybrid RGA,
Co-FS 29, 47, 48, 66, 73, 79, 81, 82, 83 GP Model-3, 85
as dry roughage, 5 least cost for dairy buffalo, 80
least cost and deviation value, hybrid RGA, least cost for cattle-1 on a DM basis, 54, 59, 62
GP Model-1, 82 least cost for cattle-2 on a DM basis, 55, 62
least cost and deviation value, hybrid RGA, least cost for cattle-3 on a DM basis, 57, 62
GP Model-2, 83 linear relationship between milk yield of
least cost and deviation value, hybrid RGA, cows and, 28
GP Model-3, 84 nutrient content of feedstuffs on a dry matter
least cost, deviation, and constraints values, (DM) basis, 48, 73
hybrid RGA, 66 nutrient requirement for buffalo, 72
least cost ration, dairy buffalo, ‘as fresh’ priority values for GP model-1, 76
basis, 81 software program developed to address, 45
least cost ration, dairy buffalo, dry matter Total mixed ration (TMR) cost
(DM) basis, 79 cutting down (reducing) while retaining
multi-cut, 47 nutrients, 60, 70
nutrient content, 6 on dry matter basis, 59, 90
nutrient content on dry matter (DM) basis, farmer’s point of view on, 85–85
48, 73 on fresh basis, 55, 56, 57, 81, 83, 84, 85, 92
Soya de-oiled cake (DOC), 47
least cost and deviation value, hybrid RGA, U
GP Model-1, 82
least cost and deviation value, hybrid RGA, Urea, 6
GP Model-2, 83 Uttar Pradesh, 2, 61
Index 113
W least cost, deviation, and constraints values,

hybrid RGA, 66
Wheat, 5
least cost ration, dairy buffalo, ‘as fresh’
Wheat bran
basis, 81
in concentrate mixtures, 28
least cost and deviation value, hybrid RGA, least cost ration, dairy buffalo, dry matter
GP Model-1, 82 (DM) basis, 80
least cost and deviation value, hybrid RGA, nutrient content, 6
GP Model-2, 83 nutrient content on dry matter (DM) basis,
least cost and deviation value, hybrid RGA, 48, 73
GP Model-3, 85 Wheat straw 73, 79, 81, 82, 83, 84
e
nc
ie
k sc
oc
st
li ve
e/
t.m
s ://
tp
ht

Livestock Ration Formulation For Dairy Cattle and Buffalo CRC Press

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Livestock Ration Formulation For Dairy Cattle and Buffalo CRC Press

Uploaded by

Copyright:

Available Formats

https://t.

Ravinder Singh Kuntal

Late Warrant Offcer R.L Verma (1956–1999)

Smt. Premvati Devi Verma

You are profoundly appreciated for your unfinching

commitment to the pursuit of my career

And also, to my son, Vivaan

and I just want to tell you that I love you so much.

1. Introduction to Animal Nutrition................................................................. 1

2. Review of Literature ...................................................................................... 9

3. Tools and Techniques................................................................................... 21

4. Binary Coded Genetic Algorithm to Solve Ration

4.2 Linear and Nonlinear Programming Model for Animal Feed

5. Least Cost Feed Formulation for Dairy Cattle During Pregnancy

6. Study of Real Coded Hybrid Genetic Algorithm to Find Least

6.4.1 Objective Function, Decision Variables, Representation

Table 5.9 Least Cost of Ration by Various Techniques for Cattle-3

Figure 4.1 Maximum Milk Yield Using GA (Original Bounds). ...................... 32

Optimization is one of the important areas of mathematics and computer science in

Shrii Saraswatii Namahstubhyam Varade Kaama Ruupini Twaam Aham

Dr. Radha Gupta earned her M.Sc. degree in mathematics from

Dr. Ravinder Singh Kuntal is an expert in optimizing and creat-

acknowledgment of Best Paper Award on “Study of Real Coded Hybrid Genetic

Dr. Vishal Patil is an expert in developing and designing least

Professor D. Rajendran, Principal Scientist, ICAR-National

1.1 Operational Research

Operations Research and the Management Sciences are the professional

1.2 Animal Nutrition

1.3 Essential Nutrients

cattle grazing in summer. Feed contains various amounts of moistures; hence,

phosphorus are calculated according to National Research Council (NRC) 2001

1.4 Classification of Feedstuffs

1.5 Application Area

In India many livestock organizations have reported shortages of feedstuff;

2.2 Linear Programming in Ration Formulation

is technically a mathematical procedure for obtaining a value weighting solution to a

Linearity: Feed formulation by LP is mainly based on the linearity between

2.2.1 Linear Programming Based Software

• It is convenient and saves manpower.

2.2.2 Limitations of Linear Programming in Ration Formulation

LP models assume nutrients levels are fxed; however, nutrient levels in

2.2.3 Overcoming the Limitations of Linear Programming

2.3 Nonlinear Programming

The primary objective of feed formulation is to generate a formula to meet the

2.4 Goal Programming

Minimize (Z) = ∑i∈m (di+ + di−) Subjected to : ∑ j=1 (aij x j + di− − di + ) = gi

where, di+, di−, xj ≥ 0 and (i = 1,2. . .m), (j = 1,2. . .n).

Minimize (Z) = ∑i∈m {di(Wi+ + Wi−)}, Subjected to : ∑ j =1 (aij x j + di + − di− ) = gi ,

2.5 Genetic Algorithm

3.1 Computation of LP Model Using MS-Excel Based Solver

3.1.1 Simplex LP Method in MS-Excel Solver

3.1.2 GRG Nonlinear in MS-Excel Solver

3.1.3 Evolutionary Algorithm in MS-Excel Solver

not depend on the derivative or gradient information; therefore, it is very diffcult

3.1.4 MATLAB (Matrix-Laboratory)

3.1.4.1 Inputs for Genetic Algorithm

3.2 Data Handling and Nutrient Bound Calculator

4.2 Linear and Nonlinear Programming Model for

4.2.1 Linear Programming Formulation

relationship, an objective function is formulated where the constraints are applied

Maximize y = 0.00403908701533 x1 + 0.25469485541324x2

4.2.2 Nonlinear Programming Formulation